JACOBS and LEVAL, Circuit Judges, with whom Judges MINER, WALKER, McLAUGHLIN, and PARKER join**:

At the close of a three-week bench trial, the United States District Court for the Southern District of New York (Motley, J.), found that defendant Vassar College ("Vassar") discriminated against plaintiff Cynthia Fisher in denying her tenure as a professor in its biology department. Fisher v. Vassar College, 852 F. Supp. 1193 (S.D.N.Y.1994). Specifically, the district court found that Vassar discriminated against the plaintiff (i) on the basis of her status as a married woman, in violation of Title VII of the Civil Rights Act of 1964, 42 U.S.C. § 2000e-2; and (ii) on the basis of her age, in violation of the Age Discrimination in Employment Act of 1967, 29 U.S.C. § 621 et seq. ("ADEA"). The district court also found that Vassar had violated the Equal Pay Act, 29 U.S.C. § 206(d)(1), by paying its junior female faculty members less than its junior male faculty members. Vassar was ordered to pay Fisher an aggregate money judgment of $626,872.12, plus attorneys' fees, and to reinstate Fisher to the rank of Associate Professor in Vassar's biology department.

A panel of this Court found the district court's ruling clearly erroneous and reversed the judgment. Fisher v. Vassar College, 70 F.3d 1420 (2d Cir.1996). During consideration of the Petition for Rehearing in Banc, a question was raised whether our review for clear error violated a rule established in Binder v. Long Island Lighting Co., 57 F.3d 193 (2d Cir.1995). A majority of the Court has decided to limit in banc review to resolution of the question whether a finding of liability under Title VII, supported by a prima facie case and a sustainable finding of pretext, is subject to review for clear error.1

No rule of law forbids appellate review for clear error in these circumstances. We hold that once an employer has proffered a non-discriminatory reason for an adverse employment action, a plaintiff in a discrimination case must show by a preponderance of the evidence that the reason for the adverse employment action was illegal discrimination. In so doing, a plaintiff may rely on the evidence constituting the prima facie case, together with supportable inferences to be drawn from the false or erroneous character of the employer's proffered reason for the adverse action. Because of the special meaning given by the Supreme Court to prima facie proof in this area of law, evidence constituting a prima facie case prior to the employer's proffer of a reason, coupled with the error or falsity of the employer's proffered reason may--or may not--be sufficient to show illegal discrimination by a preponderance of the evidence. But in any event, a finding of discrimination, like any other determination of fact, is reviewable on appeal for clear error. The panel scrutinized the district court's findings of discrimination, found them insupportable, concluded that Fisher had failed to show by a preponderance of the evidence that Vassar had discriminated against her for an illegal reason, and therefore reversed the finding of discrimination as clearly erroneous. We conclude that the panel was within its powers in reviewing the district court's finding of discrimination for clear error in these circumstances, and we therefore direct the district court to dismiss the plaintiff's suit.I

The facts of this case are fully described in the panel opinion, 70 F.3d 1420, and in the opinion of the district court, 852 F. Supp. 1193. The following brief outline of facts assumes familiarity with those opinions. Plaintiff Cynthia Fisher is a married woman who received a Ph.D. in Zoology from Rutgers University in 1963 and engaged in post-doctoral research from 1963 to 1965. From 1966 to 1974, the plaintiff devoted most of her time to raising her two children, and performed no work outside the home. From 1974 to 1976, the plaintiff took a part-time position as a lecturer in biology at Marist College.

Fisher was hired by Vassar as a visiting assistant professor in biology in 1977, and was placed in a tenure-track position in 1980. In 1982, Fisher was reappointed for a three-year term, at the end of which she was to be reviewed for tenure. During the 1984-85 academic year, Vassar undertook a comprehensive review of Fisher's candidacy for tenure. A five-member committee of tenured professors in the Biology Department, three men and two women, were charged with reviewing Fisher's credentials in accordance with four criteria: scholarship, teaching ability, leadership, and service to Vassar. In a confidential report, the committee found Fisher deficient in all four categories, and unanimously recommended that she be denied tenure. The committee's report and recommendation was forwarded to Leathem Mehaffey, chairman of the Biology Department, who informed Fisher of the committee's recommendation on March 29, 1985.

In accordance with Vassar's procedures, the departmental report and recommendation was also received by the dean of the college, the college president, and the Faculty Appointments and Salary Committee ("FASC"). (At Vassar, the dean, the president, and the FASC each make a recommendation to the Board of Trustees, the entity with the authority to grant or deny tenure.) Between April 19 and May 16, 1985, the FASC, the dean, and the Biology Department committee corresponded about Fisher's candidacy. On May 16, 1985, all five members of the FASC voted against tenure; the dean and the president concurred. Acting on these recommendations, the Vassar Board of Trustees denied Fisher tenure. In the same round of evaluations, one professor (Pinina Norrod) was granted tenure in the Biology Department, and one professor (Edward Tucker) was denied tenure. Fisher's appeal to Vassar's Faculty Appeals Committee was rejected, and she left Vassar in May 1986.

On July 7, 1987, Fisher filed her complaint in the Southern District of New York, alleging that Vassar discriminated against her on the basis of her sex. She subsequently amended her complaint prior to trial to allege discrimination on the basis of her sex conjoined with her marital status. At the close of Fisher's case, she again amended her complaint to include a claim for discrimination on the basis of age, and to add a claim under the Equal Pay Act.

After trial, the district court found that the non-discriminatory reasons proffered by Vassar for denying Fisher tenure were pretextual, a ruling that the panel opinion concluded was not clearly erroneous. The district court went on to find that the real reasons for denying Fisher tenure were discrimination based on age and on sex plus marital status. The panel opinion held that that finding was clearly erroneous, reversed and directed that the complaint be dismissed.

We have limited our in banc consideration to resolution of whether a finding of discrimination that is based on a prima facie case and a supportable finding of pretext may be reversed on appeal as clearly erroneous, or whether such a finding of discrimination must be upheld absent some quantum of evidence that the employer took the adverse action for some other non-discriminatory reason. Several well-established principles of law lead us to conclude that a finding of discrimination is reviewed for clear error like any other factual determination, and thus may be reversed--even if there is a sustainable finding of pretext--if the evidence, considered in the aggregate, will not support a finding by the district court that the reason for the adverse employment action was intentional discrimination.

We begin with the statutes. Title VII makes it unlawful "for an employer ... to fail or refuse to hire or to discharge any individual, or otherwise to discriminate against any individual ... because of such individual's race, color, religion, sex, or national origin." 42 U.S.C. § 2000e-2(a)(1). The ADEA has an analogous provision that prohibits an employer from taking an adverse employment action "because of [an] individual's age." See 29 U.S.C. § 623. The Supreme Court has held that a claim under Title VII may arise if an employer discriminates against an individual because of sex plus another characteristic, such as marital or parental status. See Phillips v. Martin Marietta Corp., 400 U.S. 542, 544, 91 S. Ct. 496, 497-98, 27 L. Ed. 2d 613 (1971) (employer discriminated against women with pre-school age children).

In McDonnell Douglas Corp. v. Green, 411 U.S. 792, 802, 93 S. Ct. 1817, 1824, 36 L. Ed. 2d 668 (1973), the Supreme Court established an "allocation of the burden of production and an order for the presentation of proof in Title VII" cases. St. Mary's Honor Ctr. v. Hicks, 509 U.S. 502, 506, 113 S. Ct. 2742, 2746-47, 125 L. Ed. 2d 407 (1993). We also apply the McDonnell Douglas framework in cases arising under the ADEA. See Woroski v. Nashua Corp., 31 F.3d 105, 108 (2d Cir.1994). Under the Supreme Court's ruling, a plaintiff alleging violation of the discrimination statutes makes out a prima facie case by showing membership in a protected class, qualification for the position, an adverse employment action, and the ultimate filling of the position by a person not of the protected class. McDonnell Douglas, 411 U.S. at 802, 93 S.Ct. at 1824; Texas Dep't of Community Affairs v. Burdine, 450 U.S. 248, 252-53, 101 S. Ct. 1089, 1093-94, 67 L. Ed. 2d 207 (1981). "The burden of establishing a prima facie case ... is not onerous." Burdine, 450 U.S. at 253, 101 S.Ct. at 1094. In fact, the plaintiff's burden of establishing a prima facie case has been frequently described as "minimal." St. Mary's, 509 U.S. at 506, 113 S.Ct. at 2746-47; Chambers v. TRM Copy Ctrs. Corp., 43 F.3d 29, 37 (2d Cir.1994) (quoting Dister v. Continental Group, Inc., 859 F.2d 1108, 1114 (2d Cir.1988)).2

"Establishment of the prima facie case in effect creates a presumption that the employer unlawfully discriminated against the employee." Burdine, 450 U.S. at 254, 101 S.Ct. at 1094. The presumption means that, unless the defendant comes forward with a non-discriminatory reason for the action complained of, the plaintiff's case may go to the jury, even though the prima facie case might be insufficient--apart from the presumption--to meet the plaintiff's ultimate burden of showing discrimination; indeed, in such a circumstance the jury must rule for the plaintiff unless the employer submits evidence that places in doubt the facts underlying plaintiff's prima facie case (such as plaintiff's qualification for the job), or furnishes a satisfactory explanation for its inability to tell the reason why plaintiff was disfavored. See St. Mary's, 509 U.S. at 509, 113 S.Ct. at 2748. "Thus, the McDonnell Douglas presumption places upon the defendant the burden of producing an explanation to rebut the prima facie case--i.e., the burden of 'producing evidence' that the adverse employment actions were taken 'for a legitimate, nondiscriminatory reason.' " Id. at 506-07, 113 S.Ct. at 2747 (quoting Burdine, 450 U.S. at 254, 101 S.Ct. at 1094). The purpose of the McDonnell Douglas framework is to force the defendant to give an explanation for its conduct, in order to prevent employers from simply remaining silent while the plaintiff founders on the difficulty of proving discriminatory intent. See Burdine, 450 U.S. at 254-56 & n. 8, 101 S. Ct. at 1094-95 & n. 8.

"It is important to note, however, that although the McDonnell Douglas presumption shifts the burden of production to the defendant, '[t]he ultimate burden of persuading the trier of fact that the defendant intentionally discriminated against the plaintiff remains at all times with the plaintiff.' " St. Mary's, 509 U.S. at 507, 113 S.Ct. at 2747 (quoting Burdine, 450 U.S. at 253, 101 S.Ct. at 1093) (alteration in St. Mary's ). Any legitimate, non-discriminatory reason will rebut the presumption triggered by the prima facie case. Thus, "[t]he defendant need not persuade the court that it was actually motivated by the proffered reasons" in order to nullify the presumption and obligate the plaintiff to satisfy the burden of proof. Burdine, 450 U.S. at 254, 101 S.Ct. at 1094; see also Board of Trustees of Keene State College v. Sweeney, 439 U.S. 24, 25 n. 2, 99 S. Ct. 295, 296 n. 2, 58 L. Ed. 2d 216 (1978) (employer need only "explain[ ] what he has done" (internal quotation marks and citation omitted)). If the defendant articulates a non-discriminatory reason, "the presumption raised by the prima facie case is rebutted, and the factual inquiry proceeds to a new level of specificity." Burdine, 450 U.S. at 255, 101 S.Ct. at 1095 (footnote omitted).

At that point, "the McDonnell Douglas framework--with its presumptions and burdens--is no longer relevant." St. Mary's, 509 U.S. at 510, 113 S.Ct. at 2749. The Supreme Court warns us not "[t]o resurrect it later":

The presumption, having fulfilled its role of forcing the defendant to come forward with some response, simply drops out of the picture. The defendant's "production" (whatever its persuasive effect) having been made, the trier of fact proceeds to decide the ultimate question: whether plaintiff has proven "that the defendant intentionally discriminated against [him]."

Id. at 510-11, 113 S.Ct. at 2749 (quoting Burdine, 450 U.S. at 253, 101 S.Ct. at 1093-94) (alteration in St. Mary's; emphasis added); see also Cabrera v. Jakabovitz, 24 F.3d 372, 382 (2d Cir.1994) ("[T]he burden of persuasion as to discrimination is on the plaintiff; the presumption that triggered the defendant's burden of production has 'drop[ped] out of the picture.' " (quoting St. Mary's, 509 U.S. at 511, 113 S.Ct. at 2749-50)); Saulpaugh v. Monroe Community Hosp., 4 F.3d 134, 142 (2d Cir.1993) ("[I]f a defendant carries its burden by producing a non-discriminatory motive, 'the McDonnell Douglas framework--with its presumptions and burdens--is no longer relevant.' " (quoting St. Mary's )). In particular, the presumption of discrimination that was raised upon a showing of the prima facie case no longer operates. See St. Mary's, 509 U.S. at 507, 113 S.Ct. at 2747; Burdine, 450 U.S. at 255-56 & n. 10, 101 S. Ct. at 1095 & n. 10 . The plaintiff then has the opportunity to demonstrate " 'that the proffered reason was not the true reason for the employment decision,' and that race was." St. Mary's, 509 U.S. at 507-08, 113 S.Ct. at 2747 (quoting Burdine, 450 U.S. at 256, 101 S.Ct. at 1095) (emphasis added). Of course, the plaintiff "retains that 'ultimate burden of persuading the [trier of fact] that [he] has been the victim of intentional discrimination.' " Id. at 508, 113 S.Ct. at 2747 (quoting Burdine, 450 U.S. at 256, 101 S.Ct. at 1095) (alteration in St. Mary's ). The question becomes the same question asked in any other civil case: Has the plaintiff shown, by a preponderance of the evidence, that the defendant is liable for the alleged conduct?

(a) The effect of a prima facie case under Title VII or ADEA. As the foregoing discussion demonstrates, the term prima facie case, as used in Title VII and ADEA actions, has a meaning that is quite different from and more limited than that ascribed to the term in many other actions. Such a limited prima facie case does not necessarily have much force in showing discrimination.

"Prima facie case" denotes what evidence a plaintiff must offer to avoid dismissal after presentation of the plaintiff's direct case. Except as to causes of actions for which special rules have been adopted, to satisfy the requirements of the prima facie case the plaintiff must present evidence from which a factfinder could reasonably find every element that the plaintiff must ultimately prove to prevail in the action. Thus, in the absence of a special policy-based rule similar to that promulgated by McDonnell Douglas, a plaintiff avoids a directed verdict only by establishing a prima facie case that assures that at the end of the trial there will be enough evidence to support a verdict in his favor (unless the defendant's evidence conclusively undermines some element of plaintiff's prima facie case).

Because of the Supreme Court's adoption of a particular framework in McDonnell Douglas and Burdine, the same is not true of a discrimination case: a plaintiff alleging discrimination can satisfy the prima facie case and avoid dismissal at the conclusion of the plaintiff's direct case without submitting evidence sufficient to support a finding in his favor on each element that the plaintiff must ultimately prove to win. The burden-shifting presumption excuses the plaintiff at that stage from showing that discrimination was present and caused the adverse employment action plaintiff suffered. If the plaintiff submits evidence of the minimal elements of the special discrimination prima facie case--membership in the protected class, qualification, adverse employment action, and preference for someone outside the protected class--the remaining elements (discrimination and causation) are presumed at this stage of the litigation, and defendant must take up the burden of going forward.

But as Burdine and St. Mary's make clear, the presumption disappears once the employer has proffered a reason. When the presumption drops away, plaintiff's burden is enlarged to include every element of the claim. Discrimination and cause are no longer presumed. To sustain the burden of putting forth a case that can support a verdict in his favor, plaintiff must then (unlike the prima facie stage) point to sufficient evidence to reasonably support a finding that he was harmed by the employer's illegal discrimination.

Accordingly, discrimination cases differ from many areas of law in that under the McDonnell Douglas burden-shifting framework a plaintiff's satisfaction of the minimal requirements of the prima facie case does not necessarily mean, even if the elements of the prima facie case go unchallenged, that plaintiff will ultimately have sufficient evidence to support a verdict on each element that plaintiff ultimately must prove to win the case.

It can be readily seen, furthermore, that the essential elements of this diminished, minimal prima facie case do not necessarily support a reasonable inference of illegal discrimination. In our diverse workplace, virtually any decision in which one employment applicant is chosen from a pool of qualified candidates will support a slew of prima facie cases of discrimination. The rejected candidates are likely to be older, or to differ in race, religion, sex, and national origin from the chosen candidate. Each of these differences will support a prima facie case of discrimination, even though a review of the full circumstances may conclusively show that illegal discrimination played no part whatever in the selection.

By this passage we do not mean to suggest that illegal employment discrimination is rare. We recognize, furthermore, that in the direct case, the evidence adduced to satisfy the prima facie standard may also amount to a powerful showing of discrimination; plaintiff's evidence of discrimination may also be powerfully strengthened by what the defendant puts forth in its case. The point we make here is that evidence sufficient to satisfy the scaled-down requirements of the prima facie case under McDonnell Douglas does not necessarily tell much about whether discrimination played a role in the employment decision. The fact that a plaintiff is judged to have satisfied these minimal requirements is no indication that, at the end of the case, plaintiff will have enough evidence of discrimination to support a verdict in his favor.

(b) The effect of a showing that the employer's reason was pretextual. A showing that the defendant's proffered reason for the adverse employment action is not the real reason may serve as evidence that the defendant intentionally discriminated. We attach the label "pretext" to a proffered reason that is not credited by the finder of fact. But the label "pretext" does not answer the question: pretext for what? In some cases, an employer's proffered reason is a mask for unlawful discrimination. But discrimination does not lurk behind every inaccurate statement. Individual decision-makers may intentionally dissemble in order to hide a reason that is non-discriminatory but unbecoming or small-minded, such as back-scratching, log-rolling, horse-trading, institutional politics, envy, nepotism, spite, or personal hostility. For example, a member of a tenure selection committee may support a protege who will be eligible for tenure the following year. If only one tenure line is available, that committee member might be inclined to vote against tenure for a junior faculty member who is currently eligible for tenure, thereby ensuring that the tenure line remains open. Any reason given by the committee member, other than the preference for his protege, will be false. Furthermore, recommenders and decision-makers who are governed by such considerations will not advise the president and regents of the institution that their recommendation or vote was disingenuous. In short, the fact that the proffered reason was false does not necessarily mean that the true motive was the illegal one argued by the plaintiff. See Wallace v. SMC Pneumatics, Inc., 103 F.3d 1394, 1399 (7th Cir.1997) (listing various non-discriminatory reasons for an employer's pretextual explanation).3

Where (as at Vassar) there are multiple recommenders or decision-makers, and where the employment decision is the result of deliberation or politicking among those recommenders or decision-makers, the problem is all the more complex. Because there are numerous participants in the decision-making process, each potentially having individual reasons for rejecting a plaintiff, there is a greater likelihood that some of those reasons will differ from the reason officially given by the institution.4

The sufficiency of the finding of pretext to support a finding of discrimination depends on the circumstances of the case. This is an unremarkable principle: the sufficiency of any evidentiary finding depends on the other findings and evidence that accompany it. What is at issue is the drawing of inferences from human behavior. Once the trial has moved to the stage at which the plaintiff must prove discrimination by a preponderance of the evidence, a defendant's false statements are nothing more than pieces of circumstantial evidence, which may be employed, as in many other types of cases, to reveal the speaker's state of mind. To the extent that an actor in defendant's position is unlikely to have proffered a false explanation except to conceal a discriminatory motive, then the false explanation will be powerful evidence of discrimination. On the other hand, if the circumstances show that the defendant gave the false explanation to conceal something other than discrimination, the inference of discrimination will be weak or nonexistent. And if, on examination of the circumstances, there are many possible reasons for the false explanation, stated or unstated, and illegal discrimination is no more likely a reason than others, then the pretext gives minimal support to plaintiff's claim of discrimination.

(c) The combined effect of a prima facie case and a finding of pretext. We have seen that, while a prima facie case and a finding of pretext may in some cases powerfully show discrimination, neither one necessarily gives plaintiff much support in discharging his obligation to prove that he was the victim of discrimination. Indeed, the combined effect of both may have little capacity to prove what the plaintiff has the ultimate burden of proving. Thus, a finding of pretext, together with the evidence comprising a prima facie case, is not always sufficient to sustain an ultimate finding of intentional discrimination. See Hargett v. National Westminster Bank, USA, 78 F.3d 836, 838 (2d Cir.) (quoting Burdine, 450 U.S. at 253, 101 S.Ct. at 1093-94), cert. denied, --- U.S. ----, 117 S. Ct. 84, 136 L. Ed. 2d 41 (1996); Sutera v. Schering Corp., 73 F.3d 13, 16 (2d Cir.1995) ("plaintiff must show that the defendant's articulated reason for its decision is in fact a pretext for discrimination"); Quaratino v. Tiffany & Co., 71 F.3d 58, 64 (2d Cir.1995) ("An employer's reason for termination cannot be proven to be a pretext for discrimination unless it is shown to be false and that discrimination was the real reason.").

A finding of pretext, therefore, does not insulate from appellate review the ultimate finding of discrimination. To permit a plaintiff to win a judgment upon a finding of pretext, without subjecting that judgment to clear error review, would impermissibly shift the burden of proof to the defendant to disprove discrimination or to offer evidence of a "third" reason. The Supreme Court has admonished, however, that "nothing in law would permit us to substitute for the required finding that the employer's action was the product of unlawful discrimination, the much different (and much lesser) finding that the employer's explanation of its action was not believable." St. Mary's, 509 U.S. at 514-15, 113 S.Ct. at 2751. Once again: "McDonnell Douglas does not say ... that all the plaintiff need do is disprove the employer's asserted reason. In fact, it says just the opposite: '[O]n the retrial respondent [plaintiff] must be given a full and fair opportunity to demonstrate by competent evidence that the presumptively valid reasons for his rejection were in fact a coverup for a racially discriminatory decision.' " Id. at 517, 113 S.Ct. at 2753 (quoting McDonnell Douglas, 411 U.S. at 805, 93 S.Ct. at 1825) (emphasis in St. Mary's ).

Accordingly, a Title VII plaintiff may prevail only if an employer's proffered reasons are shown to be a pretext for discrimination, either because the pretext finding itself points to discrimination or because other evidence in the record points in that direction--or both. And the Supreme Court tells us that "a reason cannot be proved to be a 'pretext for discrimination ' unless it is shown both that the reason was false, and that discrimination was the real reason." Id. at 515, 113 S.Ct. at 2752 (emphasis added). We have recognized again and again that a plaintiff does not necessarily satisfy the ultimate burden of showing intentional discrimination by showing pretext alone.5 A finding of pretext may advance the plaintiff's case, but a plaintiff cannot prevail without establishing intentional discrimination by a preponderance of the evidence.

The role of the appellate court is no different in reviewing a finding of discrimination than it is in reviewing any other finding of fact:

That the employer's proffered reason is unpersuasive, or even obviously contrived, does not necessarily establish that the plaintiff's proffered reason of race is correct. That remains a question for the factfinder to answer, subject, of course, to appellate review--which should be conducted on remand in this case under the "clearly erroneous" standard of Federal Rule of Civil Procedure 52(a).

Id. at 524, 113 S.Ct. at 2755-56 (citation omitted) (emphasis added). This passage is underscored by the Court's parenthetical phrase "of course," which denotes that the principle of clear error review is "certain" and "unquestionable," "as might be expected," "natural," and "obvious," and thereby furnishes a background assumption for reading other passages in the opinion. Specifically, that background assumption illuminates certain passages that (like the block-quoted passage) deal with the factfinder's ability to find discrimination once pretext is shown, but do not reiterate the caveat that such a finding is subject to the ordinary rules of appellate review as a matter "of course." Thus, elsewhere in St. Mary's, the Court states "[e]ven though (as we say here) rejection of the defendant's proffered reasons is enough at law to sustain a finding of discrimination, there must be a finding of discrimination." Id. at 511 n. 4, 113 S. Ct. at 2749-50 n. 4. Such passages do not restrict our power and responsibility to review a finding of discrimination for clear error in the usual course: even if we find that the plaintiff has made out a prima facie case and has shown pretext, we still must review a trial court's determination that the defendant intentionally discriminated for clear error.

In sum, a supportable finding of pretext and a prima facie case do not alter our ordinary standard of review for clear error, or constrain us to conclude that the defendant has intentionally discriminated:

The prohibitions against discrimination contained in the Civil Rights Act of 1964 reflect an important national policy. There will seldom be "eyewitness" testimony as to the employer's mental processes. But none of this means that trial courts or reviewing courts should treat discrimination differently from other ultimate questions of fact.

Id. at 524, 113 S.Ct. at 2755-56 (quoting United States Postal Serv. Bd. of Governors v. Aikens, 460 U.S. 711, 716, 103 S. Ct. 1478, 1482, 75 L. Ed. 2d 403 (1983)). If Binder v. Long Island Lighting Co., 57 F.3d 193 (2d Cir.1995), is read as inconsistent with this holding, we expressly reject it. Accordingly, we may reverse a district court's finding of discrimination--even if accompanied by a supportable finding of pretext--if we are firmly convinced, as the panel was here, that a mistake has been made and that the plaintiff has failed to establish intentional discrimination by a preponderance of the evidence.

Two of the three other circuits that have considered this issue have arrived at the same conclusion.6 Furthermore, of the various opinions filed herewith, none except Judge Winter's disagrees with the central proposition of this Point II--which is that a prima facie case meeting the minimal standard of McDonnell Douglas (even where elements are acknowledged by the defendant), together with a finding of pretext, do not necessarily add up to a sustainable case of discrimination.

(d) Some comments on the dissenting opinions. (i) Chief Judge Newman's dissent is based largely on his surprising view, shared by Judge Winter, that the prima facie case of discrimination specified in McDonnell Douglas is as strong as any conventional prima facie case. This position is squarely contrary to the broadly accepted view that, in discrimination cases under the law propounded by the Supreme Court, plaintiff's burden to produce a prima facie case is substantially less onerous than the conventional obligation to produce evidence that reasonably supports a finding on all the elements of the claim. If the requirements of a prima facie case under the discrimination statutes were no less onerous than is normally the case, the Supreme Court, and our Court, along with other circuits would not have said over and over again that the requirements of the prima facie case of discrimination under McDonnell Douglas are "minimal."7

Judge Newman's view comes from his highly selective misreading of the Supreme Court's opinion in Burdine. He tells us that "Burdine defined a Title VII prima facie case [as specified in McDonnell Douglas ] to mean adverse employment action taken 'under circumstances which give rise to an inference of unlawful discrimination.' " 114 F.3d at 1363 (Newman). He goes on to say, "Since those facts are sufficient to give rise to an inference of discrimination, they cannot cease to have such an effect simply because the employer has proffered an explanation." 114 F.3d at 1363.

Judge Newman's interpretation of this clause in Burdine is rebutted by the next few sentences of that opinion. Two sentences later, Justice Powell explains that "the prima facie case raises an inference of discrimination only because we presume these acts, if otherwise unexplained, are more likely than not based on the consideration of impermissible factors." Burdine, 450 U.S. at 254, 101 S.Ct. at 1094 (quoting Furnco Constr. Co. v. Waters, 438 U.S. 567, 577, 98 S. Ct. 2943, 2949-50, 57 L. Ed. 2d 957 (1978) (emphasis added, internal quotation marks omitted)). In other words, the inference wholly depends on the presumption, which disappears once the employer has proffered an explanation. The phrase "prima facie case"--as footnote 7 of Burdine says--is used in McDonnell Douglas not "to describe the plaintiff's burden of producing enough evidence to permit the trier of fact to infer the fact at issue," but rather to "denote the establishment of a legally mandatory, rebuttable presumption."8

The dissenting opinions oddly interpret this footnote to mean that the minimal prima facie case identified by McDonnell Douglas is more probative of the ultimate fact in issue than a conventional prima facie case (which, on its own, has persuasive force sufficient to support the inference that the ultimate fact in issue is more probable than not). Footnote 7, when read in combination with the text, says just the opposite.

The Burdine opinion goes on to explain, in terms reiterated in St. Mary's and closely followed in this opinion, that "[e]stablishment of the prima facie case in effect creates a presumption that the employer discriminated.... The burden that shifts to the defendant ... is to rebut the presumption of discrimination by producing evidence that the plaintiff was rejected ... for a ... nondiscriminatory reason.... If the defendant carries this burden of production, the presumption raised by the prima facie case is rebutted.... Plaintiff retains the burden of ... persuading the court that she has been the victim of intentional discrimination." Burdine, 450 U.S. at 254-56. Thus, Burdine makes clear that the presumption which bolsters the prima facie case is a transitory thing. Once the presumption has served its purpose of forcing out the employer's explanation, the prima facie case loses its capacity to support the (previously presumed) inference of illegal discrimination, other than by whatever intrinsic force the constituent evidence may have.

Finally, Burdine explains that the inference of illegal motive flows not simply from the set of facts specified in McDonnell Douglas but from those facts coupled with the absence of an explanation. Id. at 254, 101 S. Ct. at 1094 ("[T]he prima facie case raises an inference of discrimination only because we presume these acts, if otherwise unexplained, are more likely than not based on the consideration of illegal factors."). As both Burdine and St. Mary's reiterate, once the employer does what the McDonnell Douglas rule was designed to force it to do--i.e., give an explanation--the presumption disappears. Having lost the transitory benefit of the presumption, plaintiff must then satisfy the ultimate burden by showing facts from which one can reasonably find that forbidden discrimination was more probable than not. Judge Newman therefore misreads these cases when he argues that the initial, minimal McDonnell Douglas showing, on its own, supports the ultimate, sufficient finding of discrimination.9

(ii) Judge Newman argues further that the Supreme Court cannot have meant in McDonnell Douglas, Burdine and St. Mary's what we understand it to have meant because that would be unconstitutional under Mobile, Jackson & Kansas City R.R. Co. v. Turnipseed, 219 U.S. 35, 31 S. Ct. 136, 55 L. Ed. 78 (1910). In making this argument, the dissent seems once again to have misread the Supreme Court's precedent. Turnipseed, in fact, upheld the constitutionality of a statute which performed exactly the function the majority attributes to the McDonnell Douglas formula--forcing the defendant to explain. The Court stated,

The statutory effect of the rule is to provide that evidence of an injury arising from the actual operation of trains shall create an inference of negligence, which is the main fact in issue. The only legal effect of this inference is to cast upon the railroad company the duty of producing some evidence to the contrary. When that is done, the inference is at an end, and the question of negligence is one for the jury, upon all of the evidence.... The statute, does not, therefore, ... fail in due process of law, because it creates a presumption of liability, since its operation is only to supply an inference of liability in the absence of other evidence contradicting such inference.

Turnipseed, 219 U.S. at 43, 31 S.Ct. at 138 (emphasis added). Those words could serve equally well to describe the effect of the temporary inference raised by the McDonnell Douglas framework. We simply do not understand Judge Newman's argument.

All of this is admirably treated in Judge Calabresi's opinion, and we join in Part I of that opinion.

(iii) In Judge Newman's discussion of the force of a finding of pretext, he observes that our view "seems at odds" with a statement in the St. Mary's opinion that "rejection of the defendant's proffered reasons will permit the trier of fact to infer the ultimate fact of discrimination, and ... no additional proof of discrimination is required." St. Mary's, 509 U.S. at 511, 113 S.Ct. at 2749 (internal quotation marks, footnote and brackets omitted). The issues presented in St. Mary's did not require the Court to specify whether this statement ("no additional proof is required") refers to the substance of the evidence or to procedure, and the Court did not say.

We acknowledge that the sentence might be read to say that a prima facie case plus a finding of pretext is always sufficient to support a verdict in plaintiff's favor, so that such a finding precludes ordinary review for clear error. But that is not our understanding of what Justice Scalia meant.

First, if that were his meaning, he would necessarily be mistaken, as Judge Newman's dissenting opinion acknowledges. The majority and the dissent are on common ground that a finding of pretext, coupled with a prima facie case, "is not always sufficient to sustain an ultimate finding of discrimination." 114 F.3d at 1375.

Second, that interpretation would be difficult to reconcile with Justice Scalia's pointed statement at the conclusion of the opinion that the verdict is subject to review under the "clearly erroneous" standard. Finally, such a view would be incompatible with Justice Scalia's clear assertion that once the employer produces an explanation, the McDonnell Douglas framework with its presumptions and burdens no longer operates. The St. Mary's opinion emphasizes that "[t]o resurrect [the presumption] later, after the trier of fact has determined that what was produced to meet the burden of production is not credible, flies in the face of our holding in Burdine that to rebut the presumption '[t]he defendant need not persuade the court that it was actually motivated by the proffered reasons.... The presumption having fulfilled its role of forcing the defendant to come forward with some response, simply drops out of the picture.' " Id. at 510-11, 113 S.Ct. at 2748 (quoting Burdine, 450 U.S. at 254, 101 S.Ct. at 1094). See also id. at 507, 113 S. Ct. at 2747 ("If the defendant carries this burden of production [of showing a nondiscriminatory reason for the challenged action], the presumption raised by the prima facie case is rebutted ... and drops from the case.")

The main thrust of the St. Mary's majority opinion, as we read it, is that once the minimal prima facie case has served its purpose of forcing the employer to proffer a reason, all presumptions drop out and the case proceeds like any other--i.e., with the burden on plaintiff to prove the case by evidence of discrimination sufficiently persuasive to allow a favorable verdict to survive clear error review. This central meaning of St. Mary's would be contradicted if we construe the "no additional proof ... required" phrase to mean that a prima facie case plus a finding of pretext is necessarily sufficient to sustain a plaintiff's burden.

A more plausible reading of the statement that "no additional proof of discrimination is required" is that it concerns procedure. The opinion was discussing the minuet set in motion by McDonnell Douglas. For the first step, plaintiff must produce evidence to meet the minimal demands of a Title VII prima facie case. Plaintiff's doing so shifts the burden of production to the defendant to proffer a nondiscriminatory explanation. The statement that, once the defendant has made such a proffer, plaintiff is not required to produce "additional proof" of discrimination does not necessarily mean anything more than that, procedurally, the plaintiff need not proffer additional evidence but may rely on the evidence already received to rebut defendant's explanation and prove discrimination. Such a procedural observation does not diminish the plaintiff's obligation in the end to adduce sufficient evidence of discrimination to support a finding that discrimination was more probable than not.

Thus we do not think that our view of the significance of a finding of pretext is at odds with Justice Scalia's statement, properly understood.

(iv) We think Judge Winter's opinion exhibits two additional flaws of reasoning. First, Judge Winter is moved by what he sees as an anomaly--that an employer who lies can be better off than one who gives no explanation. 114 F.3d at 1390. Any apparent illogic disappears when one recognizes that the McDonnell Douglas framework is a burden-shifting rule, designed to force the employer to give an explanation. This burden shift therefore gives the plaintiff merely a transitory advantage. The employer who gives an explanation is better off than the one who declines to do so. That is no anomaly.

Second, ambiguity in the meaning of the term "prima facie case" has caught Judge Winter in a semantic trap. That term is commonly used to describe two different functions in the course of a trial. Its original meaning refers to the quantum of evidence a plaintiff must adduce to escape dismissal at the conclusion of the plaintiff's case; it is also frequently used, however, to describe the quantum of evidence which must ultimately be found in the record to justify submission of the case to the jury (or to justify upholding a jury's verdict in plaintiff's favor). The reason the same term has come to be used to describe two functionally different tests is that ordinarily the quantum of evidence needed for both purposes is the same. In the usual case, to satisfy either test there must be sufficient evidence to support a finding in plaintiff's favor on every element of the claim by a preponderance.

Judge Winter thus assumes that if the four McDonnell Douglas factors are what is needed to prevent dismissal at the close of the plaintiff's evidence, that same evidence (unless the defendant has conclusively rebutted discrimination, proved a third motive, or disproved one of the four factors) must also assure that the plaintiff has sufficient evidence ultimately to justify a plaintiff's verdict.10 But the Supreme Court has made clear that the test of sufficiency at the close of plaintiff's direct case is not the same as the ultimate test. In McDonnell Douglas, the Court referred expressly to the "initial burden ... of establishing a prima facie case." 411 U.S. at 802, 93 S. Ct. at 1824 (emphasis added). It went on to explain in language quoted above from Burdine and St. Mary's that the minimal four McDonnell Douglas factors are held to support the inference of discrimination only because of a presumption created to smoke out the employer's proffer of explanation; and that once the employer explains, the presumption drops away, and the plaintiff then must prove discrimination by evidence reasonably capable of supporting that inference. If the Supreme Court had meant what Judge Winter suggests, employers would be subjected to liability for discrimination where none was present and none was shown. We respectfully believe Judge Winter has misunderstood the Supreme Court's precedents.

* * *

Judge Newman's opinion proclaims that the majority opinion undermines the law of discrimination. We do no such thing. We have faithfully applied the law as repeatedly explained by the Supreme Court. Under that law, notwithstanding the minimal requirements of the specially defined prima facie case, once the employer has proffered an explanation, a plaintiff may not prevail without evidence that, on its own, unaided by any artificially prescribed presumption, reasonably supports the inference of discrimination.

We now turn to Fisher's claim. Fisher presented a prima facie case of discrimination by reason of her marital status by showing (i) that she was a married woman, (ii) that she was qualified for tenure, (iii) that she was denied tenure and (iv) that tenure was granted to a woman who was not married (Pinina Norrod). Fisher established a prima facie case of age discrimination by introducing the additional evidence that she was over the age of 40 at the time she was reviewed for tenure, and that eight of nine other tenured professors in the Biology Department were younger than Fisher when they were reviewed for tenure.

The burden of production then shifted to Vassar to proffer a legitimate, non-discriminatory reason for its employment decision. Vassar satisfied that burden by asserting that Fisher was denied tenure because she did not meet the posted standards for tenure, and that she was less qualified than other candidates who filled specific needs of the Biology Department. Vassar undertook to support these more general points by numerous more particularized assertions about Fisher's record. At that juncture, Fisher attempted to show that Vassar actually discriminated against her by introducing evidence that the reasons that Vassar proffered for denying her tenure were false, and by introducing other evidence intended to show that Vassar actually discriminated. This other evidence consisted of anecdotes, purported admissions made by Vassar, statistics, and expert testimony.

After an exhaustive analysis of Fisher's credentials and the credentials of other candidates for tenure in the Biology Department, the district court found that Vassar's asserted reasons were pretextual--i.e., that the reasons stated by the employer were not the real reasons for the adverse employment action. After reviewing the evidence that Fisher introduced, the district court concluded that the real reason for denying Fisher tenure was intentional discrimination. The district court found that Vassar denied Fisher tenure because of her age and her status as a married woman, and found that Vassar had violated the Equal Pay Act by paying its male junior professors more than its female junior professors. The court found against Fisher on her simple sex discrimination claim.

A panel of this Court sustained the district court's findings that Fisher had established a prima facie case of age and sex-plus discrimination. The panel also ruled that it could not find clear error in the district court's determination that certain of Vassar's assertions about Fisher's candidacy were inaccurate. But in reviewing the totality of the evidence, the panel found it insufficient to support a finding that Vassar actually discriminated against Fisher on the basis of her age or her status as a married woman. Because the panel was left with a definite and firm conviction that a mistake had been committed, it reversed the district court's decision. In light of the principles set forth in Part II above, the panel was within its power to do so.

The panel opinion employed the phrase that the finding of pretext in this case "points nowhere." The dissent makes much of this figure of speech, arguing in essence first that a finding of pretext must point somewhere and second that it points "in the same direction that all pretext findings point--toward [a] finding of discrimination." As to the first issue, the dissent is simply reading a rhetorical device with excessive literalness. What the panel opinion meant by "points nowhere" was that the inaccuracies in Vassar's statements are explicable by so many equally possible motivations that none emerged with any persuasive force; more particularly, the panel concluded that under the circumstances Vassar's inaccuracies gave little if any support to the inference that Vassar had engaged in discrimination.

We simply disagree with the dissent's suggestion that a finding of pretext in all but a few specified categories of cases reasonably supports a finding of discrimination, as the true motivation. True, a finding of pretext will in many circumstances powerfully support a finding of discrimination. In others, as explained above, it will not. Even where the proffered reason is an outright falsehood, the power of that fact as support for a finding of discrimination is not, and should not be, a rule of law but a function of logic. A finding of pretext in circumstances that suggest numerous other possible unstated explanations no less likely than discrimination gives little inferential support to a finding of discrimination.

The argument of the dissent appears to be predicated on the assumption that the laws of evidence, in the manner of the U.S. Sentencing Guidelines, assign some fixed or special value to false statements in discrimination cases. That is a misguided view. The fact that an employer has given a false explanation is a piece of evidence. Like a false exculpatory statement, its strength as an indicator of guilt will vary with the circumstances. An employer may believe that a supervisor is hiring on merit, and proffer that explanation, but a jury may find pretext nevertheless without having a basis for belief as to whether the supervisor is hiring on the basis of friendship, bribery, a hatred for 42 year-olds, or animus against Romanians--so that, even if the pretext is attributed to the employer in these circumstances, there is no evidentiary support for a finding that the employer's pretext is a pretext for discrimination. To pick up one of the dissent's analogies, flight from the scene of a crime ordinarily has evidentiary weight, but flight from a scene of arson shows nothing if the defendant fled the scene of a department store inferno during business hours. Even if the defendant does not testify to being a shopper, the bare fact of such flight does not support an inference that the defendant is the arsonist.

The dissent goes on to express disapproval of people who lie or dissemble in their testimony. But factfinding (and review for clear error) are not moral judgments; they are exercises in logic as applied to the observation of human behavior. The issue is not whether we disapprove of the defendant's lack of candor; it is whether the plaintiff has proven discrimination. If a party's conduct fails to give logical support to the finding of a fact in issue, that fact may not be found merely because we disapprove of the conduct. A court should not enter judgments for unproved, nonexistent discrimination to express its disapproval of a party's giving of inaccurate explanations in court.

To summarize, we consider how district courts should analyze discrimination cases in which the plaintiff has advanced a prima facie case, the defendant has proffered an explanation capable of being found false, and a jury has found for the plaintiff. When should such a verdict be left in place, when set aside?

The rule is that there is no rule peculiar to discrimination cases. As in all other areas, the answer depends on how forcefully the evidence has shown what plaintiff has the burden of showing--that the adverse employment action suffered by plaintiff was attributable to the alleged discrimination. If the evidence (and the inferences reasonably flowing from it) can demonstrate that the plaintiff was likely injured by the defendant's illegal discrimination, then the evidence gives rise to a jury issue, and a verdict in plaintiff's favor must be left undisturbed. If they cannot, and the jury is left to mere speculation, then the lack of evidentiary support compels the court to conclude that a mistake has been made. The court in those circumstances must find the plaintiff's verdict to be clearly erroneous and set it aside.

The point is that once the plaintiff has satisfied the minimal requirements for a prima facie case under McDonnell Douglas, and the defendant has responded by proffering a nondiscriminatory explanation, all special rules drop from the case. See St. Mary's, 509 U.S. at 507, 510-11, 113 S. Ct. at 2747, 2748-50. At this point no rule of law gives artificially enhanced weight to any piece of evidence.

Plaintiff's initial showing of discrimination may be strong or weak, depending on the particular evidence. The mere fact that a plaintiff has met the reduced demands of Title VII for a prima facie case gives little assurance that plaintiff has significant proof of discrimination.

As to the employer's proffer of a nondiscriminatory reason which the factfinder finds to be false, its probative force is also highly variable. As stated above, employers characteristically give false explanations for employment decisions for many different reasons. That an employer has done so means that there is something to hide. Discrimination is without doubt one of the things employers may seek to hide by giving a false explanation. It is by no means the only one. The fact that the employer is hiding something does not necessarily mean that the hidden something is discrimination. Generally speaking, the stronger the evidence that illegal discrimination is present, the greater the likelihood that discrimination is what the employer's false statement seeks to conceal. And, conversely, the weaker the evidence of discrimination, the less reason there is to believe that the employer's false statement concealed discrimination, as opposed to the numerous other reasons for which employers so frequently give false reasons for employment decisions.

When a court comes to consider, either upon defendant's motion for summary judgment, or after a plaintiff's verdict, whether the evidence can support a verdict of discrimination, no special rules affect the weight to be given to the prima facie case, the truthfulness or falsity of the employer's explanation, or any other piece of evidence. As in any other type of case, the judge must analyze the evidence, along with the inferences that may be reasonably drawn from it, and decide if it raises a jury question as to whether the plaintiff was the victim of discrimination. If so, summary judgment must be denied and/or a jury verdict for plaintiff must be sustained. If not, the defendant is entitled to summary judgment or to the overturning of a plaintiff's verdict as clearly erroneous.

We are sitting in banc to reconsider the question of law discussed in Part II, and not to involve all the judges of our Court in a review of the district court's factual findings. Therefore, we hereby modify our order for hearing this appeal in banc, see Fisher v. Vassar College, No. 94-7737 (2d Cir. Feb. 16, 1996) (order for rehearing) to provide that the in banc rehearing is limited to the force and effect of a pretext finding, taken together with a prima facie case, in considering on appeal whether or not an ultimate finding of discrimination is clearly erroneous under Fed.R.Civ.P. 52(a). Because our ruling on that question reaffirms the applicability of the rule employed by the panel--that no special weight is given to the evidence that supports the prima facie case and pretext in conducting clear error review of an ultimate finder of discrimination--we have no reason to suppose that the panel would now alter its views on the application of that rule to the district court's findings. Accordingly, there is no need to remand to the panel for renewed consideration in light of the in banc court's opinion.11 The mandate of the Court shall issue reversing the judgment of the district court, drawing its authority from the opinion of the in banc court as to the appropriateness of appellate review for clear error, and the panel's rulings as to the disposition of all other issues in the appeal.

The judgment of the district court is reversed. Judgment shall be in favor of the defendant.

The in banc majority, naturally, has limited its review to the legal issues presented, and therefore the majority opinion does not address Chief Judge Newman's fact-specific argument that the panel arrived at the wrong result in this case. As the author of the panel opinion, I write separately to meet the dissent on this ground.1 The best refutation of the dissent is the panel opinion; there is no complaint that it is too short, and there is no point in reprinting it. I therefore respectfully refer the reader to it, and limit myself in this separate concurrence to rebutting selected arguments and aspersions in the dissent.

* The dissent's critique of the panel's work is rooted in the dissent's errors of law. Absent "rare" and "special circumstances existing in the record," see 114 F.3d at 1372 (Newman), the dissent would insulate from reversal any ultimate finding of discrimination premised on a prima facie case coupled with a sustainable finding of pretext. Appellate scrutiny would be limited by the dissent to instances in which the record facts "overwhelmingly establish a third motive for the pretextual proffer, undermine a component of the plaintiff's prima facie case, or demonstrate that the defendant is highly unlikely to have discriminated." 114 F.3d at 1377.2 It is unclear in this passage if "overwhelmingly" modifies all three phrases or just the first, but in effect that hardly matters.

The dissent's application of its new standard of appellate review to the age discrimination claim in this case neatly demonstrates that the dissent's new rules would do away with meaningful appellate review of discrimination awards. The dissent would affirm the district court's award of more than one half million dollars under the ADEA. That award was made--after being doubled on a finding of willfulness--on the rock bottom minimum showing that the dissent argues is sufficient to withstand (or evade) appellate scrutiny for clear error.3

In determining that age discrimination had been proven, the district court found that:

(1) Fisher made out a prima facie case by the de minimis showing that she was over 40 years of age, that she was otherwise qualified for the job, and that eight of nine other tenured professors in the Biology Department were at least nine years younger than Fisher when they were reviewed for tenure.

(2) Vassar's articulated reason for denying tenure to Fisher was pretextual.

852 F.Supp. at 1230-31. There is nothing else. The district court noted that two other items of evidence were offered up to support the age discrimination claim, but the court rejected them, as the dissent observes. A review of the district court opinion confirms that there is no other evidence of age discrimination, period. The panel opinion concluded that Fisher had shown no more than that she was the oldest candidate in the biology department to have been considered for tenure, which was unremarkable because few candidates for tenure have had an eight year interruption in their careers. Moreover, Fisher's evidence consisted solely of a list of eight faculty members who were younger than Fisher when considered for tenure. The sample, too small to bear statistical weight in the first place, was selected on no evident principle. The panel therefore held that Judge Motley had committed clear error in finding that the evidence presented by Fisher amounted to sufficient proof to support an ADEA claim.

The dissent would find that the panel erred; but the dissent will not contest that the evidence of age discrimination is scant. This scant showing embarrasses the dissent's theory that a bare pretext added to a prima facie case has some special weight that evades appellate review for clear error absent a handful of barely imaginable contingencies. It is easy to see in this case that such special weight would simply be a thumb on the scale. Evidently to avoid this self-refuting application of its rule, the dissent adds to the quantum of proof a new finding to shore up its ultimate finding of age discrimination--a finding on which the district court's ADEA ruling did not rely. Thus, the dissenters dwell on Dr. Fisher's hiatus from teaching: "had Dr. Fisher not taken eight years off from teaching to raise her children, before joining the Vassar faculty, she would have been within the age group from which Vassar is apparently willing to select tenured professors." 114 F.3d at 1384.4

This is a circularity. Dr. Fisher was eight years older when she arrived at Vassar because she took an eight year hiatus. But she would have been eight years older than she otherwise was if she had devoted eight years to being a social worker (which is what Dr. Fisher did after being denied tenure), or running for office, or training for the Olympics, or serving in the military, or running a farm or business, or writing a book, or performing charitable work. Aging goes on pretty much whatever we are doing. The fact of the hiatus adds nothing to the prima facie case (which of course already factors in her age) except as a fancy way of saying that she was eight years older than she would have been if she had been eight years younger than she was. No wonder the district court did not consider this to be evidence of age discrimination. The dissent wouldn't either if there were anything else.

The language of the dissent betrays a felt need to avoid saying just where the pretext finding does point. The word "discrimination" is used by the dissent passim without specification as to the basis of any discrimination. The dissent writes:

[T]he panel opinion is quite wrong to assert that the pretext finding made by the District Court in this case "points nowhere." On the contrary, it starts out pointing in the same direction that all pretext findings point--toward the finding of discrimination that is inferable from the facts constituting the plaintiff's prima facie case.

114 F.3d at 1372-1373 (emphasis added). But this still begs the question presented to the district court and the panel: discrimination on what ground? The footnote dangling from this passage concedes that the situation is "complicated by the circumstance that the plaintiff has alleged discrimination on two different grounds." Id. at 1373 n. 8 (Newman). A further complication for the dissent is that Dr. Fisher presented a prima facie case on both grounds: age and sex-plus-marriage. But the dissent never deals with the vector of a pretext finding when--as often happens--a plaintiff alleges discrimination on three grounds, or four; where a plaintiff is actually the victim of discrimination on multiple grounds; where a plaintiff alleges two grounds of discrimination when there is no evidence that bears specifically on one more than the other; or where a plaintiff may have suffered discrimination on several grounds, some of which are not illegal. In such circumstances, the dissent offers little useful guidance for district courts in bench trials, and would require fugue-like jury charges to explain these notions to a jury.

In Dr. Fisher's case, where indeed does the pretext point? The district court found two grounds of discrimination--three of them, if one considers the bantamweight finding of sex discrimination under the Equal Pay Act. This is an embarrassment of riches to the dissent, because (unless at least one ground of discrimination is held clear error) the pretext finding must point in multiple directions.

Different courts and judges have different ideas about where (if anywhere) the pretext points in this case. (A) The district court held that the evidence points to willful discrimination on the basis of Dr. Fisher's age as well as on the basis of her status as a married woman.5 (B) The dissenters agree with the district court that the pretext points decisively to discrimination against the aged and against married women. The dissenters feel, however, that the pretext points "more strongly" to sex-plus discrimination; but they conclude that it was "within the prerogative" of Judge Motley to infer discrimination from the pretext on both grounds because "marital status is not realistically incompatible with additional discrimination based on age." 114 F.3d at 1385. Thus, for the dissent, pretext (out of doctrinal necessity) points in as many directions as can be deemed consistent.6 (C) Judge Calabresi's opinion asks whether Vassar's assertion that Fisher spent insufficient time in the laboratory could be premised on a stereotypical view that married women with children "spend less time in the lab." 114 F.3d at 1360. It could--if there were any evidence that Vassar's decision-makers were in the grip of such a stereotype. But marriage would seem to be beside the point; the pretext finding that Judge Calabresi posits could point as well to discrimination based on a general stereotype of women as care-givers, and directed against all working women, or (perhaps) against all working mothers. (D) Finally, the panel concluded that the district court's pretext finding pointed nowhere.

The most that one can say about the direction of the pretext finding in this case is that, like a compass at the Pole, it wheels all over the place.

Having shown that the dissent's deferential review of the district court's age discrimination finding is all deference, no review, I turn to what the dissent believes is evidence that points to and supports an ultimate finding of discrimination on the basis of Dr. Fisher's status as a married woman.

It is obvious that any employment decision reflects a preference for the candidate who gets the spot. Here, Dr. Pinina Norrod won tenure in the Biology Department the same year that Dr. Fisher lost. The district court found that:

... the senior members of the Biology Department were determined that Dr. Fisher be denied tenure and that Dr. Norrod be promoted and granted tenure.

852 F.Supp. at 1217 (finding # 92). The dissent adopts the operating assumption that there are only two potential motives for that preference: (a) merit ("a college [is] usually regarded as a bastion of uninhibited pursuit of truth," 114 F.3d at 1370), or (b) illegal discrimination. That is an invalid assumption for the creation of inferences. Accepting the district court's finding that the tenure selection process was skewed to prefer Dr. Norrod, the question becomes: what evidence offers an explanation for that preference? Dr. Norrod had been married at the time she was hired by Vassar. T. at 2283. Although she had been divorced at the time the tenure decision was made, she was then in a long-term relationship that led to marriage soon after. T. at 2283-84; 70 F.3d at 1475. Was Vassar skewing its tenure decisions in the Biology Department, as the district court held, to get rid of Dr. Fisher because she was married in order to grant tenure to another woman because she was between marriages? In any event, does this finding command affirmance, as the dissent would hold, without further scrutiny for clear error?

The dissent, 114 F.3d at 1383 (Newman), draws confidence on this point from the district court's finding that "the Biology Department's senior faculty" had accepted a "stereotype and bias: that a married woman with an active and on-going family life cannot be a productive scientist and, therefore, is not one despite much evidence to the contrary." 852 F. Supp. at 1216 (finding # 87). In reading this finding, one should keep in mind that the chairman of Vassar's Biology Department for nine years had been Dr. Patricia Johnson, a mother who raised her daughters in that time. T. at 2283. The district court dealt with that inconvenient fact by drawing the contours of the supposed prejudice so as to exclude Dr. Johnson: the operative words in finding # 87 are "married" and "on-going," which excise (the divorced) Dr. Johnson from the picture--a bit of "stereotype and bias" to the effect that a divorced mother does not have "an active and on-going family life." For the same reason, the wording of finding # 87 excludes Dr. Norrod, the woman who (between marriages) won tenure in 1985 over Dr. Fisher as well as over Dr. Edward Tucker.

The district court also limited its survey to married women in the "hard" sciences (finding # 80), and narrowed the field further still by excluding the married women who were granted tenure in the Psychology Department, on the ground that psychology is deemed to be a "soft" science (notwithstanding undisputed testimony that Vassar's Psychology Department is considered part of the hard sciences due largely to its focus on experimental laboratory studies) (T. at 2216-18).

Finding # 87 reflects Dr. Fisher's strategy at trial. She defined the ground of discrimination by reference to a set of characteristics that describe her, but that are so particular as to create in a small field a category of one.7 Following Dr. Fisher's lead, the district court arrived at the conclusion that there is a sex-based animus at work at Vassar that denied tenure to Dr. Fisher so that preferment could be given to women who have divorced, women who have not yet married, and married female scientists who work in psychology rather than biology. The dissent would hold that this inference enjoys a special procedural insulation from ordinary clear error review.

The dissent also fixates upon a couple of specific findings made by the district court, to which I turn now.

A. Teaching Evaluations.

As to the panel's review of the statistical evidence and anecdotal evidence in the case, the dissent "for purposes of this discussion, ... assume[s]" that the panel was right. 114 F.3d 1382 (Newman). But the dissent points to a group of findings that supposedly demonstrate "discriminatory use of student evaluations." Id. I will review those findings as briefly as possible. After the district court found (finding # 47) that Vassar students evaluate teachers on a scale of one to five, and (finding # 48) that Dr. Fisher exceeded what the Dean of Faculty called a "good statistic" of 70 fours and fives combined, the court went on to find that the tenure review committee:

reached its unfavorable conclusions as to Dr. Fisher's teaching record by applying different standards to her than were applied to other tenure candidates. The Committee distorted the numbers by counting only fives for Dr. Fisher while counting fours and fives for the other candidates to determine student evaluation rankings. (R. at 289-99, 1146.)

(finding # 49 (emphasis added)). Taken at face value, this finding supports the conclusion that Vassar did not give Dr. Fisher a fair shake.8 The dissent points to these findings as weighty evidence of sex-plus discrimination, but they do not at all suggest that Dr. Fisher was disfavored because she had not divorced her husband; and they are entirely consistent with the Department's preference for Dr. Norrod, a preference that could have numerous grounds, differing from one faculty member to the next.

But the dissent errs in accepting finding # 49 at face value, and in reading it (as the language lends itself to being read) to mean (1) that the tenure review committee used a numerical bench-mark--a combined score of fours and fives--to measure the relative teaching ability of tenure candidates; (2) that "other candidates" were measured by this benchmark; and (3) that Dr. Fisher was downgraded because her score--distorted by counting fives only--was lower than the teaching score of other candidates.9 These impressions do not survive a look at the underlying evidence.

The district court gives two record references to support finding # 49: "R. at 298-99, 1146." 852 F. Supp. at 1209. The first record cite is to Dr. Fisher's own testimony, testimony that says nothing about how the college used or did not use the student evaluation numbers; in that passage, Dr. Fisher testified about charts prepared for trial by her own counsel tabulating the numerical results of the student evaluation sheets for Dr. Fisher and the other specified candidates. These charts arguably support the idea that Dr. Fisher was a good teacher (hence the adequately supported finding of pretext), but they do not show how if at all the college used this data.

The second record cite ("1146") is to the testimony of the plaintiff's expert, Dr. John Castellot, a professor of biology at Tufts University. Dr. Castellot never taught at Vassar and was not involved in the tenure review process; he conceded that he had never served on a tenure review committee anywhere. R. 1155. Indeed, his testimony was offered primarily to show that Dr. Fisher's research in the field of biology compared favorably with the research of her colleagues. Over Vassar's objection, Dr. Castellot testified as to the tenure review process at Vassar as follows:

In looking over the data collected on teaching, what struck me was that the tenure committee appears to only consider the percentage of fives in the top of the scale when looking at Dr. Fisher's teaching evaluations, whereas they consistently lumped fours and fives together, which is what we do at our institution--at least what I do in my courses.

* * * * * *

It struck me that for Dr. Fisher there was frequent mention of only the fives, without including the fours together. The same standard was not applied to the other individuals that were being considered for tenure. In this case, fours and fives were lumped together.

R. 1164. In this passage, Dr. Castellot apparently was referring to two sets of documents. First, he referred to three "outside evaluations" of Dr. Fisher performed for the Biology Department by biologists at other institutions (Exhs. 12A, 12B and 12C). These evaluations do not refer to the one-to-five numerical ratings, and on the whole are favorable to Dr. Fisher.

Second, Dr. Castellot referred to the tenure assessment letters prepared by the Biology Department for Drs. Suter, Mehaffey, Hemmes, Norrod, and Fisher (Exhs. 323, 304, 292, 338, and 216). These documents do not indicate or suggest that the teaching ability of tenure candidates was measured by any single formula or benchmark. They show that, in assessing teaching ability, the tenure committee considered far more than just student evaluations, and gauged teaching ability based on various factors such as course difficulty and the introduction of new courses to the curriculum. Most important, the tenure assessment letters employ no uniform statistical method in evaluating the teaching skills of tenure candidates, and did not use any disparate method in evaluating Dr. Fisher. Sometimes--for Dr. Fisher and others--the letters consider the combined number of fours and fives; sometimes the letters consider fives only--particularly if the number of fives reflect some remarkable feature, such as all fives for a course, or none, or (as in Dr. Fisher's case) a fall-off over time from many fives to none. The particulars from the record are set forth in the margin.10

The district court relied on finding # 49 (and the related findings) chiefly to show pretext and Dr. Fisher's qualifications for tenure. The dissent relies on these findings to show that the reason for disfavoring Dr. Fisher's candidacy was sex-plus-marriage discrimination. These findings do not begin to support that inference. Dr. Castellot's "opinion"--phrased in terms of "what struck me"--is unsupported by evidence of any detectable difference in treatment between Dr. Fisher and her colleagues in terms of the use of the student evaluations. And, as the dissent (barely) concedes, that weight is arguably "lessened" by finding # 55, which says and explains why "[a]ny comparison to Dr. Norrod's teaching evaluations, which were excellent, does not give a true picture of the situation." 852 F. Supp. at 1210. Thus, according to the district court, the teaching evaluations of Dr. Norrod--the only successful candidate who competed head to head with Dr. Fisher for tenure in 1985--cannot be usefully compared with Dr. Fisher's evaluations. I cannot think why the dissent--in light of that--would place determinative weight on the teaching evaluations in order to explain why Dr. Norrod was chosen and Dr. Fisher was not.

B. Hiatus.

The dissent identifies a second set of findings as lending great support to an ultimate finding of sex-plus discrimination: the emphasis on Dr. Fisher's hiatus from her profession. Dr. Fisher completed her post-graduate studies in 1966. From then until 1974, she devoted herself to her family full-time. From 1974 to 1977, she taught part-time at Marist College. Her first full-time employment in biology was at Vassar in 1977. In the tenure evaluation process, some faculty members mentioned--and some emphasized--that Dr. Fisher had been out of the field for ten years. The dissent characterizes these findings as part (actually, close to all) of the "strong" case of sex-plus discrimination, and chides the panel for discounting them.

The dissent argues that consideration of such a hiatus, absent an adverse impact on a person's current attainments, unfairly burdens "working women." 114 F.3d at 1383. I agree. However, "working women" is a category that takes in all women in the workplace and therefore bears upon simple sex discrimination, rather than sex-plus-marriage. See Coleman v. B-G Maintenance Management of Colorado, Inc., 108 F.3d 1199, 1203 (10th Cir.1997) (to prove sex-plus-marriage discrimination plaintiff must show that subclass of married women were treated differently than subclass of married men). As it happens, however, the district court rejected Dr. Fisher's claim of sex discrimination as baseless. 852 F. Supp. at 1225.

The dissent is much more avid than the district court about the weight of the hiatus issue. The dissent deems the hiatus comments "further evidence" of discrimination on the basis of sex-plus-marriage. 114 F.3d at 1384. But the district court, after noting the Biology Department's position (expressed at a faculty committee meeting) that Dr. Fisher was "[o]ut of date--out of the field for 10 years," found: "these statements do not establish, by themselves, direct evidence of an illegitimate factor." 852 F. Supp. at 1231 n. 25. (As discussed above, the dissent deems the hiatus to be probative of age discrimination as well, though the district court did not.)

In any event, evidence of such a hiatus could have bearing only on a sex-based disparate impact claim. However, as the panel opinion and the district court opinion emphasize, no disparate impact claim was pleaded or mounted here. See 70 F.3d at 1443, 852 F.Supp. at 1226 n. 15. Thus, although this issue might have potential for pleading and development under a theory of disparate impact, there is no basis for considering that claim in this case.

CALABRESI, Circuit Judge, concurring in part and dissenting in part.

There are, it seems to me, four principal issues that divide this court. The first is the weight to be given to the plaintiff's prima facie case of discrimination once the defendant has proffered an explanation for its actions. The second is the significance to be afforded to the fact that the explanation is found to be pretextual. The third is whether, given the in banc court's answer to the first two questions and the history of this case, it is (a) appropriate, (b) permissible but unwise, or (c) simply illegitimate for the in banc court to leave standing the original appellate panel's conclusions. The fourth and final point is whether--applying the in banc court's answer to the first two questions to the facts of this case--a reversal of the trial court's finding of discrimination is warranted. (Only the dissenters, Judge Jacobs, and--by implication--the original panel members, have spoken to this last issue).

On the first two questions, I agree with the majority of the in banc court. As to the third, I agree with the dissenters' conclusions (though only as a matter of prudence rather than of propriety). Finally, I am not yet prepared to reach a final decision on the fourth issue, and think it unwise to do so now.

What is the significance, once the defendant has proffered an explanation, of the fact that the plaintiff has, admittedly, made out a Burdine- St. Mary's prima facie case of discrimination? See Texas Dep't of Community Affairs v. Burdine, 450 U.S. 248, 252-54, 101 S. Ct. 1089, 1093-94, 67 L. Ed. 2d 207 (1981); St. Mary's Honor Ctr. v. Hicks, 509 U.S. 502, 506, 113 S. Ct. 2742, 2746-47, 125 L. Ed. 2d 407 (1993). On this point, I join Parts II(a), II(c), and II(d) of the majority opinion. I add a few words because I believe that the dissenting opinions are unduly reductionist in their view of presumptions and prima facie cases, and that Chief Judge Newman's opinion in particular misconstrues those ancient Supreme Court railroad accident cases, Mobile, Jackson, & Kansas City R.R. Co. v. Turnipseed, 219 U.S. 35, 31 S. Ct. 136, 55 L. Ed. 78 (1910) and Western & Atl. R.R. v. Henderson, 279 U.S. 639, 49 S. Ct. 445, 73 L. Ed. 884 (1929). Properly understood, I believe these cases support the majority on this issue.

In Turnipseed, the Court found constitutional a state statute that permitted an inference of negligence to be drawn against the railroad defendant from the fact of a railroad accident. While requiring a rational connection between the fact shown (the accident) and the fact inferred (negligence), the Court found that such a link existed. It neither required nor found that reasonable people would believe more probably than not that the existence of a railroad accident bespoke railroad negligence. It simply found that the correlation between railroad accidents and railroad negligence was high enough to justify--constitutionally--the placement of the burden of explaining what happened on the railroad. Indeed, the connection was strong enough, the Court held, to allow a state to require a verdict against the railroad if it failed to answer. See Turnipseed, 219 U.S. at 43-44, 31 S.Ct. at 137-38.

Henderson did not overrule Turnipseed. It held, however, that the link between a railroad accident and railroad negligence, while strong enough to allow the placing of the burden of response on the railroad under Turnipseed, was not strong enough to survive and affect the jury's determination once the railroad had proffered an answer. This was because reasonable jurors could not find that the fact to be proven (the railroad's negligence) followed, more probably than not, from the fact demonstrated (the accident). In other words, the correlation between the two was high enough to impose on the defendant the duty to answer, but it was not strong enough to support a jury verdict after the defendant had proffered that answer. See Henderson, 279 U.S. at 642-44, 49 S.Ct. at 447-48.

Turnipseed and Henderson do not fit within the dissenters' tidy schema. The inference in Turnipseed is, in Judge Winter's terms, a strong prima facie case. That is, it compels (rather than simply permits ) a judgment for the plaintiff if it is not answered. Turnipseed, 219 U.S. at 43, 31 S.Ct. at 138. And yet, it is sufficiently weak--it does not reflect strong enough correlations between the fact shown and the fact to be inferred--to survive and be considered by the jury once the defendant has answered. See Henderson, 279 U.S. at 642-44, 49 S.Ct. at 447-48.2

Applying Turnipseed and Henderson to the situation before us, what do we see? We see that certain facts that correlate with discrimination have been deemed sufficient to require an answer from the defendant. That is the meaning of Title VII as read by Burdine and St. Mary's. We also see that over time the facts the plaintiff must show to require such an answer have become relatively minor. Case after case emphasizes their de minimis nature. See, e.g., Chertkova v. Connecticut Gen. Life Ins. Co., 92 F.3d 81, 90 (2d Cir.1996); Chambers v. TRM Copy Ctrs. Corp., 43 F.3d 29, 37 (2d Cir.1994).3 In other words, though still sufficiently correlated with discrimination to satisfy a requirement that the defendant answer on pain of losing, these facts, like the facts of an accident in Turnipseed that also required an answer, do not necessarily, and without more, support a finding of discrimination. Indeed, if Title VII required that facts so weakly correlated to discrimination as those we have deemed sufficient to make out a prima facie case, be enough--in the presence of an answer by the defendant--to support a finding of discrimination, Title VII would give rise to significant constitutional doubts under Henderson and its progeny.

It will not do to say, as Chief Judge Newman's dissent seems to, that since Henderson requires a strong relationship between the fact shown and that inferred, plaintiff's case necessarily meets that requirement. To the contrary, Henderson quite clearly distinguished between what is necessary to require an answer and what is necessary to permit a factfinder, without more, to find discrimination. On the one hand, the Henderson Court did not disturb Turnipseed's holding that a statute was valid to the extent that it stated that an accident and railroad negligence were sufficiently linked to require an answer by the railroad. On the other hand, the Henderson Court struck down the Georgia statute insofar as that statute, on the basis of the same correlation, permitted a factfinder to find that there was railroad negligence whenever there was an accident.

Under the circumstances, there are only two ways out. We may follow the majority and say that all that the plaintiff's prima facie case does is what the valid statute did in Turnipseed--require the defendant to speak. And once the defendant has spoken, the existence of sufficient evidence of discrimination must be determined anew by looking at all the evidence independently from the existence of the previous temporary inference/presumption. Alternatively, we may reject the majority's view and declare that a plaintiff may not make out a prima facie case under Title VII without presenting evidence that is, in the first instance, sufficiently correlated with discrimination so that reasonable people can find more probably than not that there is discrimination whenever the plaintiff has made such a showing. This interpretation would, in my judgment, force us to demand, for a prima facie case, evidence of discrimination that goes well beyond the de minimis amount that the McDonnell Douglas factors, as accepted by Burdine and applied consistently by the circuit courts, have deemed sufficient.

The dissenters disagree. They believe that Burdine, in effect, limits the role played by the McDonnell Douglas factors by providing that those factors are enough to establish a prima facie case only in "circumstances which give rise to an inference of unlawful discrimination," and that in such circumstances they are sufficiently correlated with discrimination to meet the requirements of Henderson. I think that the dissenters are wrong.

In the first place, Burdine defines "circumstances which give rise to an inference of unlawful discrimination" by giving the four McDonnell Douglas factors as an example. It makes clear that those factors are sufficient (but not necessary) to establish what it calls a prima facie case. See Burdine, 450 U.S. at 253-54 & n. 6, 101 S. Ct. at 1093-94 & n. 6. Accordingly, Burdine cannot fairly be interpreted to add any requirements for a prima facie case beyond what was demanded in McDonnell Douglas, and what it means by "inference" must be a Turnipseed inference. Indeed, after Burdine, courts have continued regularly to rely on the four McDonnell Douglas factors to determine when a de minimis prima facie case exists.

In the second place, while the dissenters criticize the phrase de minimis, they accept the description that the requirements are " 'not onerous.' " 114 F.3d at 1366, n. 3 (quoting Burdine, 450 U.S. at 253, 101 S.Ct. at 1093), or "minimal" 114 F.3d at 1366, n. 3 (quoting St. Mary's, 509 U.S. at 506, 113 S.Ct. at 2746-47). And in this case they find that Fisher has made out a prima facie case of age discrimination on what in actuality is little more than a simple McDonnell Douglas showing that: 1) she was over forty years old; 2) she was qualified; 3) she was rejected; and 4) the position was given to another applicant with similar qualifications.4 Yet the dissenters then assert that this showing is sufficiently correlated with discrimination to meet the Henderson test. In my opinion, it demonstrably is not.5 Whether the dissenters do this because they believe that this showing really does point sufficiently to discrimination, or whether, instead, they are confused by the fact that this showing has been termed a "prima facie case," and that this phrase in other areas of the law implies evidence sufficient to sustain a jury verdict, is not for me to say. But if I am right that it isn't sufficiently correlated, then what they wish to do is forbidden by Henderson, and neither failing to call the factors de minimis, nor calling what the factors establish a "prima facie case" or an "inference," makes them right.6

Both of the ways out that I have suggested have problems. Making the requirement of a prima facie case strong enough so that it truly supports, without more, a fact-finding of discrimination, would, in my judgment, do two things that no one in the in banc court has recommended: (1) require the overturning of Burdine and McDonnell Douglas as well as a large number of decisions in this and other circuits; and (2) undercut the eminently reasonable requirement that defendants in Title VII cases be made to explain their actions relatively frequently.7 But saying that a prima facie case is not enough to support a factfinding of discrimination has--as the dissents point out--its own problems. Mainly, it requires us to say that the term "prima facie case" is used differently in Title VII than elsewhere. And this is unpleasant. It is, nevertheless, the better solution.

It would have been nice if Turnipseed and Henderson had said that the plaintiffs there had made out a prima facie case of negligence sufficient to require the defendants to answer, but insufficient to support a jury verdict once an answer was given. They used different words instead. (Their words--conflating inferences and presumptions--had their own problems.) But the point they made, and which the majority makes, is nonetheless the same, and it is a profoundly wise one. There are situations in which it makes sense to force the defendant to explain itself and its actions, once a plaintiff has shown facts which, though too weak to demonstrate more probably than not that the defendant should be liable, are nevertheless sufficiently linked to possible wrongdoing by the defendant to justify an explanation. In the end, whether one says that such a showing by the plaintiff gives rise to an inference, a temporary presumption, or a prima facie case (or, for that matter, Thucydides or Mustard Plaster) is relatively unimportant. What counts is that we not let verbal confusions keep us from the result that Title VII--as long interpreted by the courts--seeks to achieve: that the plaintiff can force an explanation out of the defendant with relative ease, but that the defendant can only be held liable (once an answer is given) when a finding of discrimination is truly supported by the facts. The majority stays true to that course, and I therefore join it in this respect.

What significance is to be given to the fact that the explanation given by the defendant is pretextual? One might have thought that the requirement of an answer would mean that an honest answer was required, and that absent an honest answer the defendant should be treated as having given no answer at all. That position was, of course, rejected by St. Mary's. See St. Mary's, 509 U.S. at 509, 113 S.Ct. at 2748. (Indeed, that is the only one of the many important statements in St. Mary's that is holding and not dictum.) Some may find that holding paradoxical. There is, however, a plausible explanation for it,8 and in any event, we are bound to follow it.

I cannot doubt that the existence of a pretextual explanation, without more, points in the direction of discrimination. It does not weaken, and must at least minimally, and potentially forcefully, strengthen the evidence of discrimination that the plaintiff originally adduced. It was in part because the panel seemed to deny this that I, and I believe others as well, voted to rehear this case in banc.9 On this issue, unlike the issue of the prima facie case, there is now at least verbal agreement, among all members of the in banc court. (That agreement may, however, mask a division on how much weight in practice should be given to a pretextual answer.) The members of the panel have joined an opinion stating that their original comment that a pretext finding "points nowhere" was a rhetorical device, 114 F.3d at 1346, and the dissenters agree that the importance of a pretextual explanation varies from virtually nil, in some circumstances, to great in others, 114 F.3d at 1342-1344. All in all, I find the current discussion of pretext in Part II(b) of the majority opinion satisfactory, and so I join it. I add a few words of explanation, however, because they will serve to explain, to some degree, my disagreement with the majority on points three and four.

I begin by making explicit what is implicit in both the majority opinion and in Chief Judge Newman's dissent--that the strength of the pretext, and therefore the amount of evidence required to counter it,10 depends on context. This is, in part, because pretexts come in many varieties. Consider the following two examples:Example 1: A university says that it denied someone tenure because she did not publish enough articles. It turns out that this is a lie--she published as many as those promoted, and the university knew it.

Example 2: A university says that it denied someone tenure because she was too aggressive to fit in. It turns out that this is a lie--males promoted were manifestly as aggressive.

The first example points to discrimination, but weakly. It does so only because a lie suggests an ulterior motive, and in the context of a claim brought--and a prima facie case made--under Title VII, the existence of any ulterior motive is some evidence of discrimination. The second example offers stronger evidence of discrimination. It is not only a lie, but a lie linked to a discriminatory stereotype--namely, that women are and ought to be less aggressive than men. Many other examples of even stronger and weaker pretexts could be given, and all pretexts can be seen as arrayed on a continuum ranging from those that provide the weakest support for a finding of discrimination to those that provide the strongest support.

Under St. Mary's, any pretext permits (but does not require) a finding of discrimination that will be upheld on appeal unless such a finding is clearly erroneous.11 See St. Mary's, 509 U.S. at 524, 113 S.Ct. at 2756. But whether a finding is clearly erroneous necessarily depends on the nature of the pretext as well as on what other evidence of discrimination has been presented. Where the pretext points strongly to specific discrimination, the existence of a "third reason" or clear evidence of a lack of discriminatory intent may well be required to counter the inference of discrimination. Where the pretext points only weakly or indirectly toward discrimination, evidence of non-discrimination may be much more general. Such evidence may, for example, amount to no more than hints of possible third reasons together with the existence of a generally non-discriminatory attitude or history on the part of the employer. And this would be especially so if the prima facie case for discrimination is--as it can frequently be in Title VII cases--minimal.

Based on this discussion of pretext and the preceding one of the prima facie case, we can formulate a standard for the weight to be ascribed to each. One could summarize the relevant standard--in the first instance for the trier of fact but subject to the normal rules of review--by making seven points. First, all pretexts point toward a discriminatory motive and, if uncontroverted, tend to support (but do not require) a finding of discrimination. Second, all pretexts may be countered by sufficient evidence of non-discrimination. Third, all pretexts do not give equal support to a finding of discrimination. Fourth, the quality and quantity of the evidence deemed sufficient to counter a pretext will depend on the pretext's nature--weak or general evidence will suffice to overcome a pretext that gives rise only to a weak inference of discrimination, while strong and specific evidence will be required to counter a pretext that gives rise to a strong inference. Fifth, what evidence is sufficient depends also on the strength of the prima facie case and of the other evidence for discrimination that has been introduced. Sixth, it is therefore erroneous to believe that all pretexts can be overcome only by certain types of "strong" evidence (e.g. specific evidence of third reasons or specific evidence of a mistaken belief that the plaintiff was in the allegedly favored category). This is particularly true because of the de minimis nature of the prima facie case in the employment discrimination context. Seventh, quite apart from whether the pretext is countered or not, once the defendant has answered the plaintiff's prima facie case by giving an explanation, even if that explanation is pretextual, the plaintiff's case stands on its own and can be tested for sufficiency by an appellate court. The fact, in other words, that the plaintiff has made out a Burdine- St. Mary's prima facie case, and that the defendant has strengthened that case by giving a pretextual explanation, may, but need not necessarily, mean that a factfinder's determination of discrimination can stand.III.

The majority modified the order granting this rehearing in banc to provide that the rehearing was "limited to the force and effect of a pretext finding, taken together with a prima facie case, in considering on appeal whether or not an ultimate finding of discrimination is clearly erroneous under Fed.R.Civ.P. 52(a)." 114 F.3d at 1347. Because it found that the in banc rulings on these issues of law were consistent with the rules employed by the panel, the majority deferred to the panel's assessment of the facts. While I do not believe that the majority had an obligation to engage in its own independent assessment of the facts, I think that it would have been prudent for it to do so. This is so because I find the question quite close on the facts, and also because I sense that an engagement with the facts might have clarified at least part of the disagreement between the majority and Chief Judge Newman's dissent on the weight to be given, in practice, to a finding of pretext.

Assuming we accept the principles set forth in Parts I and II of this opinion, we still have to decide whether in this case the evidence of non-discrimination is sufficient to counter the suggestion of discrimination that arose from Vassar's dissembling, and more generally whether taking the plaintiff's and the defendant's evidence as a whole, Judge Motley's finding of discrimination is clearly erroneous. I find this question close, in part because the strength of the suggestion of discrimination raised by Vassar's lie is less easily gauged than in many cases. Vassar's pretext could be viewed as just a lie, or as a lie that derives from a discriminatory stereotype. Vassar incorrectly claimed, for example, that Fisher did not spend enough time in the lab. Was this a simple lie or was it based on a stereotypical view that married women with children spend less time in the lab?12 I also find the question close because--like the panel--I believe the district court erred in some of its findings, and I am unsure of the full significance of those errors.

Under the circumstances, I therefore agree with the dissenters' conclusion that the in banc court should not defer to the original panel's determinations. I emphasize, however, that I do so only on the grounds of prudence. The deference accorded to the panel by the in banc court is perfectly legitimate--I simply think it is unwise.

The final question is whether a reversal of the trial court's finding of discrimination is warranted. Were I the finder of fact, and were I forced to decide at this stage of the proceedings, I think that I would not have found that the plaintiff had met her burden of proving discrimination. I believe the prima facie case is very weak, that little additional evidence of discrimination has been presented, that the pretext is "semi-strong," and that there is sufficient generalized countervailing evidence to overcome both the pretext and the other evidence supporting discrimination. At the same time, applying the ordinary rules of review for clear error, I am far from sure that I would find the district court's determination of discrimination to be clearly erroneous.

Either conclusion would require a far deeper examination of the record than I have made. I do not need to make that examination, however, given the majority's decision to defer to the panel. I agree with the panel that at least in some instances the district court erred in how it went about finding discrimination. But I do not know whether, once these errors are corrected, the district court on remand would or would not find that the plaintiff met her burden of proving discrimination. Thus, I am not ready to assume, as Chief Judge Newman's dissent seems to do, that sending the case back for a reconsideration free of errors would necessarily lead to the same result. 114 F.3d at 1347. Indeed, I believe that such a view gives insufficient credit to the distinguished district judge who has heard this case. But I am also not ready to conclude, as did the panel (and, in effect, the majority), that--cleansed of error--there would be insufficient evidence of discrimination, and therefore that any finding of discrimination would require reversal on appeal.

For these reasons, I would have the in banc court either itself undertake a thorough examination of the alleged errors of the district court or remand the case to the panel for such an examination.13 Thereafter, the district court would be instructed to reconsider its holding, in the light of these errors. Finally, I would put off any determination as to whether the ultimate finding of the district court was sustainable under clear error review, until I had before me its new and reconsidered decision (whatever that might be).

Accordingly, while I concur in Parts II(a), II(b), II(c), II(d), and IV of the majority opinion, I would vacate the judgment of the district court, and hence respectfully dissent from the in banc court's reversal of that judgment and from its award of judgment to the defendant.

JON O. NEWMAN, Chief Judge, with whom Judges KEARSE, WINTER, and CABRANES concur, dissenting:

The result of this rehearing in banc is the entirely unwarranted rejection of a trial judge's ultimate findings of discrimination against married women and age discrimination in the denial of tenure to a college professor, even though those ultimate findings are supported by facts establishing a prima facie case of discrimination on both grounds, by a sustainable finding that the employer's proffered reason for denying tenure was pretextual, and by additional sustainable findings pointing to discrimination. This result is accomplished by a combination of actions of the in banc court and the panel that originally decided this appeal. Though I do not doubt that, in some Title VII and ADEA cases, an ultimate finding of discrimination may properly be ruled clearly erroneous despite a valid finding of pretext, there is no basis for such a ruling on this appeal.

I respectfully dissent, not only because of disagreement with the outcome of this appeal, but, more significantly, because the combined rulings of the in banc majority and the panel depart from settled law and practice in several respects concerning discrimination law, appellate review, and in banc practice. One of the most bizarre aspects of the majority's opinion is its insistence that the in banc court is deciding only a very narrow issue even as the majority opinion advances significant views on matters far beyond the narrow issue identified. The majority says that the in banc court is deciding only the issue of "whether a finding of liability under Title VII, supported by a prima facie case and a sustainable finding of pretext, is subject to review for clear error." 114 F.3d at 1333. That issue was never in question, since findings of discrimination had always been subject to "clear error" review, and the Supreme Court had confirmed the availability of such review, prior to this in banc rehearing, in St. Mary's Honor Center v. Hicks, 509 U.S. 502, 511, 113 S. Ct. 2742, 2749-50, 125 L. Ed. 2d 407 (1993). Beyond reasserting the obvious availability of "clear error" review, the in banc majority undertakes to express several views concerning the significance (or lack of significance) of a prima facie case, a finding of pretext, and the combination of both in the context of discrimination cases.

Understanding the reasons for my disagreement with the majority requires separate consideration of (1) what the majority has said about discrimination law, especially the significance of a prima facie case and the significance of a finding of pretext, (2) the role of an appellate panel in reviewing the factual findings of a district court, (3) the role of an in banc court in relation to the panel whose decision is being reheard, and (4) the panel's consideration of the merits of Dr. Fisher's claims of discrimination. Since these issues relate primarily to Dr. Fisher's claim of discrimination against her as a married woman, I consider them in that context, and discuss thereafter her age claim.

A. The Claim of Discrimination Against Married Women

1. Discrimination Law and the Significance of a Prima Facie Case and a "Pretext" Finding

(a) The Significance of a Prima Facie Case. The majority asserts that a "prima facie case" under Title VII, assessed without regard to the persuasive force of any opposing evidence, is not necessarily sufficient to take the ultimate issue of discrimination to the fact-finder. In the majority's view, a Title VII prima facie case serves only to require the defendant to proffer an explanation for its adverse action; upon lack of such a proffer, the plaintiff is entitled to judgment as a matter of law. But, says the majority, if the defendant proffers an explanation, the plaintiff's prima facie case is not necessarily sufficient to take the case to the fact-finder, i.e., to support an inference of discrimination. Though I recognize that the facts of a plaintiff's prima facie case might be so undermined or so overwhelmed by opposing evidence that no reasonable fact-finder could find discrimination, I disagree with the view that the facts of a prima facie case, assessed without regard to opposing evidence, are insufficient to support an inference of discrimination.

The phrase "prima facie case" has long been recognized to have two meanings. It usually means evidence sufficient to permit (but not require) a fact-finder to find a disputed matter in favor of the party presenting the prima facie case. This is the prima facie case that creates a permissible inference. In some limited circumstances, it can mean, in addition, evidence sufficient to require a finding of a disputed matter in favor of the party presenting the prima facie case in the absence of a proffer of evidence by the opposing party. This is the prima facie case that creates a rebuttable presumption. The Supreme Court made it clear in Texas Department of Community Affairs v. Burdine, 450 U.S. 248, 101 S. Ct. 1089, 67 L. Ed. 2d 207 (1981), that in the Title VII context, the Court was using the phrase "prima facie case" to mean evidence that creates a rebuttable presumption. See id. at 254 n. 7, 101 S. Ct. at 1094 n. 7.

But, as Judge Winter's dissenting opinion points out, the prima facie case that creates a rebuttable presumption is an even stronger version of the concept of a prima facie case than the version that only permits an inference. Moreover, the Supreme Court in Burdine defined a Title VII prima facie case to mean adverse employment action taken "under circumstances which give rise to an inference of unlawful discrimination."1 Id. at 253, 101 S. Ct. at 1093-94 (footnote omitted). The four subsidiary facts given in McDonnell Douglas Corp. v. Green, 411 U.S. 792, 802, 93 S. Ct. 1817, 1824, 36 L. Ed. 2d 668 (1973), as an example of a prima facie case of discrimination are not four isolated bits of information plucked from the air. They are the Supreme Court's own example of facts sufficient, in the Supreme Court's words, to "give rise to an inference of unlawful discrimination." Burdine, 450 U.S. at 253, 101 S.Ct. at 1094 (emphasis added). Since those facts are sufficient to give rise to an inference of discrimination, they cannot cease to have such an effect simply because the employer has proffered an explanation. That proffer defeats the presumptive effect that the four facts would have had in the absence of an explanation (i.e., the plaintiff is no longer entitled to judgment as a matter of law once the four subsidiary facts have been proven), but it does not deprive these facts of their capacity to support an inference of discrimination if the fact-finder finds these facts proven and then chooses to draw such an inference. The Supreme Court explicitly made this point in Burdine:

A satisfactory explanation by the defendant destroys the legally mandatory inference of discrimination arising from the plaintiff's initial evidence. Nonetheless, this evidence and inferences properly drawn therefrom may be considered by the trier of fact on the issue of whether the defendant's explanation is pretextual.

Burdine, 450 U.S. at 255 n. 10, 101 S. Ct. at 1095 n. 10. There is simply no way to reconcile the Supreme Court's statement in Burdine that the four facts of a prima facie discrimination case are facts sufficient "to give rise to an inference of unlawful discrimination" and the majority's assertion that "[t]he fact that a plaintiff is judged to have satisfied these minimal requirements [of a prima facie case] is no indication that, at the end of the case, plaintiff will have enough evidence of discrimination to support a verdict in his favor." 114 F.3d at 1337 (emphasis added).

The capacity of facts supporting a rebuttable presumption to permit the inference of the ultimate fact even after the opposing side has proffered an explanation was explicitly recognized by Congress in adopting Rule 301 of the Federal Rules of Evidence, a rule the Supreme Court cited in Burdine in explaining a Title VII rebuttable presumption. See Burdine, 450 U.S. at 255 n. 8, 101 S. Ct. at 1094-95 n. 8. Rule 301 provides:

In all civil actions ... a presumption imposes on the party against whom it is directed the burden of going forward with evidence to rebut or meet the presumption, but does not shift to such party the burden of proof in the sense of the risk of nonpersuasion, which remains throughout the trial upon the party on whom it was originally case.

Fed.R.Evid. 301.

The Conference Report explaining this Rule states:

Under the Senate amendment, a presumption is sufficient to get a party past an adverse party's motion to dismiss made at the end of his case-in-chief. If the adverse party offers no evidence contradicting the presumed fact, the court will instruct the jury that if it finds the basic facts, it may presume the existence of the presumed fact. If the adverse party does offer evidence contradicting the presumed fact, the court cannot instruct the jury that it may presume the existence of the presumed fact from proof of the basic facts. The court may, however, instruct the jury that it may infer the existence of the presumed fact from proof of the basic facts.

The conference adopts the Senate amendment.

H.R. Conf. Rep. No. 93-1597, at 2 (1974), reprinted in 1974 U.S.C.C.A.N. 7098, 7099 (second emphasis added).

Indeed, if the facts constituting a Title VII prima facie case could not permit an inference of discrimination (and in the majority's view they sometimes do not), then it is very likely that these facts could not constitutionally have the presumptive effect that McDonnell Douglas and Burdine hold that they have in the absence of a defendant's proffered explanation. The Supreme Court's rebuttable presumption jurisprudence long ago established that a presumption of "one fact from evidence of another" will satisfy due process requirements only if "there shall be some rational connection between the fact proved and the ultimate fact presumed, and that the inference of one fact from proof of another shall not be so unreasonable as to be a purely arbitrary mandate." Mobile, Jackson & Kansas City R. Co. v. Turnipseed, 219 U.S. 35, 31 S. Ct. 136, 55 L. Ed. 78 (1910). Turnipseed considered the constitutionality of a state statute providing that "proof of injury inflicted by the running of locomotives or cars of [a railroad] company shall be prima facie evidence of the want of reasonable skill and care on the part of the servants of the company." Id. at 41, 31 S.Ct. at 137. The Court understood the statute to cast upon the railroad the burden of producing some evidence disputing its negligence, without which the defendant would lose as a matter of law. The Court upheld the constitutionality of the statute because "its operation is only to supply an inference of liability in the absence of other evidence contradicting such inference" and because there was "some rational connection between the fact proved and the ultimate fact presumed." Id. at 43, 31 S. Ct. at 138.

Though some might view Turnipseed as a case that permitted a rebuttable presumption that was sufficient to impose liability in the absence of a defendant's proffer of opposing evidence, but was created by facts insufficient in themselves to carry a case to a fact-finder, i.e., the fact of injury and the fact that the injury was inflicted by a train, the Supreme Court thought otherwise: "It is not an unreasonable inference that a derailment of railway cars is due to some negligence, either in construction or maintenance of the track or trains, or some carelessness in operation." Id. at 44, 31 S.Ct. at 138.2

The requirement of Turnipseed, that it must be reasonable to infer the presumed fact from the proven facts, was subsequently invoked to invalidate rebuttable presumptions in a series of criminal cases, beginning with Tot v. United States, 319 U.S. 463, 63 S. Ct. 1241, 87 L. Ed. 1519 (1943). Id. at 467-68, 63 S.Ct. at 1244-45. Tot, it will be recalled, involved a statute that made possession of a firearm by a convicted felon "presumptive evidence" that the firearm was transported or received in interstate commerce by the felon. That the Court viewed the presumption as rebuttable merely by the proffer of opposing evidence is made plain by the Court's statement that the statute "leaves the jury free to act on the presumption alone once the specified facts are proved, unless the defendant comes forward with opposing evidence." Id. at 469, 63 S.Ct. at 1245-46. This rebuttable presumption was invalidated as in conflict with due process because, in the absence of a proffer of opposing evidence (which need not have been proven), it permitted inference of an ultimate fact that was not rationally connected to the facts that were proved.

The Supreme Court subsequently applied Tot to invalidate rebuttable presumptions in United States v. Romano, 382 U.S. 136, 86 S. Ct. 279, 15 L. Ed. 2d 210 (1965) (rebuttable presumption of control of still from presence at still), Leary v. United States, 395 U.S. 6, 29-54, 89 S. Ct. 1532, 1544-57, 23 L. Ed. 2d 57 (1969) (rebuttable presumption of illegal importation of marijuana from possession of marijuana), and Turner v. United States, 396 U.S. 398, 418-19, 90 S. Ct. 642, 653-54, 24 L. Ed. 2d 610 (1970) (rebuttable presumption of illegal importation of cocaine from possession of cocaine). Leary explained that rebuttable presumptions are valid only where "it can be said with substantial assurance that the presumed fact is more likely than not to flow from the proved fact on which it is made to depend." Leary, 395 U.S. at 36, 89 S.Ct. at 1548 (footnote omitted). Leary also made clear that the Court's "more likely than not" test for rebuttable presumptions was not influenced by the context of a criminal case in which guilt must be proved beyond a reasonable doubt. The Court said that the rebuttable presumption was invalid for failure to pass the "more likely than not" test without reaching the further question of whether the presumption would be valid to meet the "reasonable doubt" standard. Leary, 395 U.S. at 36 n. 64, 89 S. Ct. at 1548 n. 64.

Since a rebuttable presumption is constitutionally valid only where "it can be said with substantial assurance that the presumed fact is more likely than not to flow from the proved fact," Leary, 395 U.S. at 36, 89 S.Ct. at 1548, it necessarily follows that the rebuttable presumption that the Supreme Court applied to Title VII cases in McDonnell Douglas and Burdine is based on subsidiary facts from which it can be said with substantial assurance that the presumed fact of discrimination is more likely than not to flow. Surely, the Supreme Court did not create in McDonnell Douglas and Burdine a rebuttable presumption that was unconstitutional under Turnipseed, Tot, and Leary. It also necessarily follows that the facts that suffice to create the McDonnell Douglas /Burdine rebuttable presumption are facts from which a reasonable trier may infer discrimination. Since these facts will support a constitutionally valid rebuttable presumption only if it is more likely than not that the ultimate fact--discrimination--flows from the subsidiary facts, it must be reasonable for a trier to draw the inference of that ultimate fact. Thus, the facts that support the rebuttable presumption in Title VII cases must also be sufficient to create a prima facie case in the lesser and ordinary sense that they suffice to permit a fact-finder to find the ultimate fact of discrimination in the plaintiff's favor.

Brushing aside the caselaw on rebuttable presumptions, the majority nonetheless asserts that "prima facie case" in Title VII has a third meaning, one far weaker than the only two identified by the Supreme Court in Burdine, 450 U.S. at 254 n. 7, 101 S. Ct. at 1094 n. 7. In the majority's view, a Title VII "prima facie case" starts out meaning facts sufficient to create a rebuttable presumption and then, upon the proffer of an explanation by the defendant, becomes a set of facts that only sometimes permits a fact-finder to infer discrimination. Though the majority does not say that the facts establishing a prima facie case of discrimination are always insufficient to get to a fact-finder, it does assert that since such a prima facie case is relatively easy to present, such facts, though supported by evidence, will sometimes be insufficient even to create a factual issue as to discrimination. In the majority's words:

[A] plaintiff alleging discrimination can satisfy the prima facie case and avoid dismissal at the conclusion of the plaintiff's direct case without submitting evidence sufficient to support a finding in his favor on each element that the plaintiff must ultimately prove to win.

114 F.3d at 1336-37 (emphasis in original). The majority has thus embraced a novel concept--the insufficient prima facie case.

The majority views a prima facie discrimination case as a collection of facts that always suffices to pry an explanation out of a defendant on pain of suffering an adverse judgment if no explanation is proffered and only sometimes also suffices to permit a jury to infer discrimination. The Supreme Court has never so characterized a prima facie discrimination case. It is true that the Court has characterized the burden of establishing a prima facie of discrimination as "not onerous," Burdine, 450 U.S. at 253, 101 S.Ct. at 1093-94, but it has not said that facts constituting a prima facie case, if not undermined by opposing evidence, are insufficient to permit an inference of discrimination. In Burdine, the Court said precisely the opposite.

St. Mary's made the further point that, after a defendant has proffered an explanation and thereby removed the presumptive force of the plaintiff's prima facie case, the prima facie case does not compel a directed verdict in the plaintiff's favor. St. Mary's, 509 U.S. at 515, 113 S.Ct. at 2751-52 (emphasis added). That is because the fact-finder must decide whether or not to draw the inference of discrimination from the facts constituting the prima facie case. But though St. Mary's quite understandably values the prima facie case at less than what is required to obtain a directed verdict, it does not reduce it so far as to say, as the majority here holds, that a prima facie case, even where its constituent facts are credited, sometimes might not even get to a fact-finder for decision. Thus, the Supreme Court is content to say that, after a defendant proffers an explanation, even a false one, a prima facie case does not guarantee the plaintiff a victory. The majority here says that a supportable prima facie case does not even guarantee the plaintiff an opportunity to have the fact-finder consider the case.

Some of the confusion in this area of the law stems from the way the Supreme Court has articulated the McDonnell Douglas/Burdine analysis and the lower courts have applied it. McDonnell Douglas identified four factors that were present in that case and sufficed to constitute a prima facie case. The sentence identifying the four factors carried an important footnote, which is often overlooked. It says:

The facts necessarily will vary in Title VII cases, and the specification above of the prima facie proof required from [the plaintiff] is not necessarily applicable in every respect to differing factual situations.

McDonnell Douglas, 411 U.S. at 802 n. 13, 93 S. Ct. at 1824 n. 13.

Then in Burdine the Court generalized the test for a prima facie case by stating that the plaintiff "must prove by a preponderance of the evidence" adverse action taken "under circumstances which give rise to an inference of unlawful discrimination." Burdine, 450 U.S. at 253, 101 S.Ct. at 1094. Again, the Court added a footnote of explication. It referred to the four factors that had sufficed in McDonnell Douglas as "an appropriate model for a prima facie case of racial discrimina tion," explained again that "this standard is not inflexible," and quoted from footnote 13 of McDonnell Douglas the important language that " '[t]he facts necessarily will vary in Title VII cases, and the specification above of the prima facie proof required from [the plaintiff] is not necessarily applicable in every respect in differing factual situations.' " Burdine, 450 U.S. at 253 n. 6, 101 S. Ct. at 1094 n. 6 (quoting McDonnell Douglas, 411 U.S. at 802 n. 13, 93 S. Ct. at 1824 n. 13).

These were important qualifications to be relegated to footnotes, and some courts might not have always kept them in mind. Thus, some courts repeat the four factors that were given in McDonnell Douglas as an example appropriate for that case and seem to regard them as always enough for a prima facie case, even though the different fact patterns such courts are confronting might not meet Burdine 's general test of "circumstances which give rise to an inference of unlawful discrimination." And some courts, including ours, have said that the prima facie case may be "de minimis," a phrase not normally associated with a prima facie case sufficient to take the ultimate issue to a fact-finder.3 However, even when we have labeled a prima facie case "de minimis," we have usually, perhaps always, done so with respect to a fact pattern that met Burdine 's standard of "circumstances which give rise to an inference of unlawful discrimination." See, e.g., Chambers v. TRM Copy Centers, 43 F.3d 29, 37 (2d Cir.1994) (collecting cases upholding plaintiff's prima facie case).

Focusing on the four factors, without regard to the precise facts of a case, is what has led the majority to believe that we in the dissent favor subjecting employers "to liability for discrimination where none was present and none was shown." 114 F.3d at 1344. The majority attributes that view to us because they assume that we think a prima facie case of discrimination is established whenever the four factors illustrated in McDonnell Douglas are met, regardless of what facts the plaintiff presents to establish the prima facie case. That is not our view. We take seriously the caution in McDonnell Douglas, repeated in Burdine, that the facts "necessarily will vary in Title VII cases," and that the "specification" of "prima facie proof required" in McDonnell Douglas "is not necessarily applicable in every respect to differing factual situations." McDonnell Douglas, 411 U.S. at 802 n. 13, 93 S. Ct. at 1824 n. 13.

The majority mischaracterizes our view in saying that we think an employer may be found liable without evidence of discrimination. We believe that a prima facie case requires, as the Supreme Court said in Burdine, adverse employment action taken "under circumstances which give rise to an inference of unlawful discrimination," Burdine, 450 U.S. at 253, 101 S.Ct. at 1094, and that if such a prima facie case is presented, and if, at the end of the trial, the inference remains reasonable despite the defendant's evidence, then, as the Court said in St. Mary's, "no additional proof of discrimination is required," St. Mary's, 509 U.S. at 511, 113 S.Ct. at 2749 (brackets and emphasis omitted), and the case goes to the fact-finder for decision. The facts in McDonnell Douglas sufficed to support such an inference; a qualified Black was rejected and the employer sought other workers with no better qualifications.4 The facts also sufficed in Burdine, where a qualified woman was rejected, and after several months, the position was filled by a male who had been under her supervision. Burdine, 450 U.S. at 254 n.6, 101 S. Ct. at 1094 n.6. The facts will not necessarily suffice every time a member of some protected class is rejected. The precise facts, not the rote repetition of the four factors, determine whether a prima facie case, sufficient to support an inference of discrimination and thus to permit a finding of liability, has been presented.

In any event, I believe that the Supreme Court wishes to recognize a prima facie case under Title VII only when the facts presented support a reasonable inference of discrimination. Whatever confusion in such terms as "inference" and "presumption" might have existed when the Court decided Henderson in 1929, the Court that decided Burdine in 1981 and St. Mary's in 1993 understood that an inference permits, but does not require, finding an ultimate fact from a predicate fact or facts.5 That is why a Title VII prima facie case can exist only when the predicate facts support a reasonable inference of discrimination. As the Court has cautioned, however, the fact patterns that meet that standard will vary from case to case.

Of course, a prima facie case of discrimination might cease to have its normal effect of permitting a jury to infer discrimination in some circumstances. But these instances can arise only where (a) the evidence in the entire record is such that no reasonable fact-finder could find to be true the facts presented by the plaintiff as constituting the prima facie case (e.g., the evidence indisputably shows that the plaintiff is unqualified), or (b) the evidence in the entire record points so strongly away from discrimination (or toward a third reason) that no reasonable fact-finder could infer discrimination. The elimination of a prima facie case in such circumstances, either by being indisputably undermined or indisputably overcome, can occur only where evidence provides a basis for such conclusions. The prima facie case, sufficient in the Supreme Court's view to permit an inference of discrimination, cannot, in the absence of undermining or opposing evidence, cease to permit the inference just because a majority of this Court believes, without reference to such evidence, that the inference ought not to be drawn.

(b) The Effect of a Finding of Pretext. Having embraced a new and weak meaning of "prima facie case," the majority then compounds its error by substantially downgrading the effect of a finding of pretext. In this case, the trier of fact found that Vassar's proffered reason for denying Dr. Fisher tenure was a pretext. The proffered reason, as acknowledged by the majority, was that Dr. Fisher "did not meet the posted standards for tenure, and that she was less qualified than other candidates who filled specific needs of the Biology Department."6 114 F.3d at 1345. The panel opinion concluded that the District Court's finding of pretext was sustainable, Fisher v. Vassar College, 70 F.3d 1420, 1437 (2d Cir.1995) ("Fisher II "), and the majority opinion, by disclaiming any consideration of the panel's assessment of any of the District Court's findings, leaves undisturbed the critical panel ruling that the pretext finding is not clearly erroneous.

Some of what the in banc majority says about a pretext finding is not in dispute. We all agree that a finding of pretext permits the fact-finder to draw an inference of discrimination. As the Supreme Court explicitly stated, "[R]ejection of the defendant's proffered reasons will permit the trier of fact to infer the ultimate fact of intentional discrimination, and ... no additional proof of discrimination is required." St. Mary's, 509 U.S. at 511, 113 S.Ct. at 2749 (internal quotation marks, footnote, and brackets omitted; emphasis in original). We also agree that the fact-finder need not infer discrimination after finding the proffered reason to be a pretext. The inference is available, not required.

We begin to part company when the majority offers its analysis of what a pretext finding means. The majority begins its analysis by pointing out that a pretextual reason for justifying adverse employment action might be advanced for a variety of reasons. The majority correctly observes that decision-makers might "intentionally dissemble" to hide such true reasons as "back-scratching, log-rolling, horse-trading, institutional politics, envy, nepotism, spite, or personal hostility." 114 F.3d 1337. Such reasons, the majority points out, are non-discriminatory. To this unexceptional starting point, the majority then adds the equally indisputable point that the probative force of a pretext finding on the ultimate issue of discrimination is not a fixed value, but varies depending upon the circumstances of each case.

The majority then introduces three considerations to justify its view that the pretext finding in this case may properly be given slight, if any, probative force. Each consideration risks weakening the law of discrimination, at least in the context of university tenure decisions. First, the majority relies on the large number of reasons why a pretext explanation could be proffered as a circumstance for diminishing the force of the pretext finding that was made by the District Court. But numbers of possibilities have little to do with probative force, as can be readily demonstrated by consideration of a common instance of a permissible finding of state of mind--the inference of consciousness of guilt from flight from the scene of a crime. A person running from a crime scene might have done so for many reasons. He might be late for a doctor's appointment, or a date with a friend, or any other important event; he might have just remembered that he left the oven on in his apartment, or the shower running; he might have heard a loud noise and thought there was some explosion endangering him; or he might just be a jogger who runs around that block every morning. The majority apparently believes that whenever a number of possible explanations can be suggested for conduct that normally has probative significance, the probative force of the conduct diminishes, triers of fact should be wary of attaching significance to it, and appellate courts should be emboldened to reject as clearly erroneous an ultimate finding that relies in part on a normally available inference drawn from such conduct.

This view of the law is totally unsupported and unsupportable. In every case where a person is seen running from the scene of a crime, the jury is instructed that they may, but need not, draw an inference that such flight is probative of consciousness of guilt, and that state of mind, in turn, is probative of ultimate guilt (though not alone sufficient to convict). Of course, the strength of the inference from flight will depend on all of the evidence in the case. If there is some slight evidence of an innocent explanation for the flight, the inference of consciousness of guilt might be lessened. If the evidence of an innocent explanation is strong, the inference of consciousness of guilt will be weak. And cases might arise where the evidence of an innocent explanation is so strong that no reasonable fact-finder could draw the inference of consciousness of guilt, in which event a finding based on the inference would be clearly erroneous.

To assert these self-evident propositions is not to suggest, as the majority says of this dissent, 114 F.3d at 1345, that I favor assigning a "fixed" or "special" value to a proffer of a pretextual explanation for an adverse employment decision. I do not. I simply observe, as every appellate court that has ever had a pretext case has observed, that a finding of pretext will normally permit (not require) an ultimate finding of discrimination, that the possibility that other explanations can be imagined is not a valid reason for depreciating the significance of a pretext finding, and that an appellate court cannot validly point to such possibilities as a basis for declaring clearly erroneous a trial court's decision to infer discrimination from a finding of pretext.

The majority then amplifies its argument about multiple possible explanations by stating that a pretext finding is to be accorded "minimal" significance "if, on examination of the circumstances, there are many possible reasons for the false explanation, stated or unstated, and illegal discrimination is no more likely a reason than others." Id. at 1338 (emphasis added). This statement is offered in an effort to provide an explanation of what the panel meant by its extraordinary statement that the District Court's pretext finding "points nowhere." See Fisher II, 70 F.3d at 1437. I can agree that a fact-finder 's determination that discrimination is no more likely to be the motive for the proffer of a pretext than many other explanations is an ample basis for according diminished weight to a pretext finding. But the majority's gloss on the "points nowhere" statement provides no basis for depreciating the pretext finding in this case. That is so because in this case, there was nothing in the record to support a finding that log-rolling, nepotism, spite, or any of the majority's other hypothesized explanations were the reason for Vassar's proffer of a pretextual explanation, and the fact-finder did not find that other explanations for the pretextual proffer were as likely as discrimination. Whether the majority's "equal probability" gloss and its unflattering speculation about how tenure decisions are made at Vassar strengthens the argument for ruling the discrimination finding clearly erroneous on appellate review is more properly considered in Part A(2), infra.

The majority's second reason for diminishing the significance of a pretext finding is the startling notion that an employer like Vassar might quite understandably proffer a false reason for adverse employment action simply because of a "lack of candor." 114 F.3d at 1346. Thus, a college, usually regarded as a bastion of uninhibited pursuit of truth, is excused from proffering its real reasons because the professors holding them lack "candor."

However, the whole point of the three-part analysis of St. Mary's and its predecessors (plaintiff's prima facie case, defendant's proffer, and plaintiff's ultimate burden to prove discrimination) is to afford the defendant an opportunity to proffer its real reason for the challenged action. A Title VII lawsuit is not some sort of face-saving exercise to enable bashful decision-makers to excuse their lack of candor. The defendant is called upon to proffer what it believes is the true reason for its action. It need not do so precipitously. It can make appropriate inquiry within its ranks. But once it proffers its reason in court, it subjects that reason to the assessment of the fact-finder, and if the fact-finder concludes, with support in the record, that the proffered reason is a pretext, the defendant is usually at risk of having the fact-finder draw the permissible inference that the pretextual reason was proffered to hide the true reason--discrimination.

Third, the majority depreciates the significance of a pretextual explanation by observing that the tenure decision challenged by Dr. Fisher was the combined result of many decision-makers. For several reasons, this circumstance is also not a proper basis for depreciating the force of a pretext finding.

To begin with, Vassar did not say, after full opportunity for internal inquiry, that different participants in the tenure process held views related to "log-rolling," "envy," "spite," or any of the other reasons suggested as possibilities by the majority. Vassar said Dr. Fisher was denied tenure because she lacked tenure qualifications, either on an absolute basis or at least on a comparative basis, in view of the needs of the Biology Department. Having proffered its single explanation, Vassar cannot expect the probative force of a finding that this explanation is pretextual to be diminished just because Vassar could have proffered a variety of explanations that might have been held by its various decision-makers.

Moreover, whenever motivation is in issue, as it is in every discrimination case, identification of motive is more difficult where decisions are made by a group than by an individual. That fact, however, cannot provide insulation for group decision-making permeated by discrimination. Perhaps if only one out of 100 decision-makers participating in a collective process harbored an impermissible motive, the group decision would not be vulnerable. But if some significant portion of the decision-makers shared the improper motive (or perhaps just one of a very small number), the decision is tainted. The Supreme Court has recognized this principle in ruling that a state constitutional convention acted with an impermissible, racially-based motive in enacting a constitutional provision, even though the evidence identified only some convention delegates holding that motive. See Hunter v. Underwood, 471 U.S. 222, 228-30, 105 S. Ct. 1916, 1920-21, 85 L. Ed. 2d 222 (1985) (citing evidence canvassed by court of appeals in Underwood v. Hunter, 730 F.2d 614, 618-20 (11th Cir.1984)); cf. United States v. O'Brien, 391 U.S. 367, 383-84, 88 S. Ct. 1673, 1682-83, 20 L. Ed. 2d 672 (1968) (declining to declare act of Congress unconstitutional based on motives expressed by "a handful of Congressmen"). Whatever approach courts might take in ascertaining the motives of a group of convention delegates or legislators for purposes of constitutional adjudication, the group decision-making process of an employer, tested for purposes of determining a Title VII violation, may be found to be tainted by an impermissible motive held by any significant participants in the process. The Supreme Court made this precise point in Anderson v. City of Bessemer City, 470 U.S. 564, 579-80, 105 S. Ct. 1504, 1514-15, 84 L. Ed. 2d 518 (1985), when it upheld the reasonableness of the District Court's finding of discrimination by citing the pretextual explanations of just two members of the five-member committee that had rejected the application of the Title VII plaintiff.

A further point about group decision-making concerns the venerable principle of respondeat superior. If an employer entrusts personnel decisions to persons who act on the basis of impermissible motives, the employer is responsible for the resulting act of discrimination. It may be true, as the majority says, that Vassar is not lying about the reason for denying tenure to Dr. Fisher just because one or more professors in the tenure process are lying, or as the majority prefers to say, "intentionally dissembl[ing]." Nevertheless, an employer can no more avoid legal responsibility for the unlawful discriminatory action of its subordinates than it can for their negligent infliction of harm in the course of their employment.

For all of these reasons, I fundamentally disagree with the majority's view that the pretext finding in this case, or in other tenure denial cases, is generally to be accorded slight evidentiary significance on the ultimate issue of discrimination, or, in the panel's even more extreme view, "points nowhere," Fisher II, 70 F.3d at 1437, at least in the absence of evidence that blunts the probative force of the pretext finding. In my view, the proper analysis as to the significance of a pretext finding should start, not with excuses for an employer's lack of candor nor with a claim that such a finding "points nowhere," but with an understanding of what a "pretext" really is. A standard dictionary definition is "a purpose or motive alleged ... in order to cloak the real intention." Webster's Third New International Dictionary 1797 (1993). Thus, a pretext is not merely a reason that relies on factual assertions that are not true; it is a reason that a person proffers but does not believe is the real reason for the action the person has taken. If a supervisor believes an employee has embezzled funds, a discharge based on that belief is not rendered pretextual simply because at trial the employee proves that he did not embezzle. The supervisor's proffered reason for the discharge is a pretext only if the supervisor says that embezzlement was the reason for the discharge and does not believe that this was the real reason. Of course, evidence that the employee did not embezzle is probative of the supervisor's true state of mind, and a trier persuaded that the employee did not embezzle might well doubt the claimed statement of belief.7 But the issue remains--is it true that the supervisor believed that the employee embezzled? The reason stated ("proffered" in Title VII parlance) is pretextual only if the person does not believe what he or she is saying. What must be proffered is the alleged reason for the challenged action, i.e., why the action was taken.

In ordinary discourse, if a person asserts that something is true and knows that it is not, that person would be described as lying (or, in the majority's felicitous phrase, "intentionally dissembl[ing]"), at least in the absence of joking, rhetorical debate, or mental disorder. But not every misstatement of fact, even a misstatement as to one's own belief, is necessarily a lie, in the sense of an indictable perjury. Though the distinction between a knowingly false statement of one's belief and a lie is not easy to describe, the majority in St. Mary's drew this very distinction. See 509 U.S. at 520-21, 113 S.Ct. at 2754-55. The distinction inheres in this significant sentence from Justice Scalia's opinion:

The factfinder's disbelief of the reasons put forward by the defendant (particularly if disbelief is accompanied by a suspicion of mendacity ) may, together with the elements of the prima facie case, suffice to show intentional discrimination.

Id. at 511, 113 S.Ct. at 2749 (emphasis added). Since the falsity of the reason (e.g., the employee did not embezzle or was never late) is not necessarily inconsistent with an employer's belief that the reason is true, in which case the employer's motive is not impermissible, Justice Scalia must have been using the phrase "disbelief of the reasons" to mean a fact-finder's "disbelief that the defendant really believed the reasons it was putting forward." That being so, Justice Scalia must have meant that mendacity is something additional to a defendant's putting forward a reason that it knows is not true.

Justice Scalia offers two explanations as to why a pretextual reason is not necessarily a lie, at least in the sense of a criminally actionable perjury. First, he points out, a dispute between two versions of an occurrence underlying the employer's proffered reason, even though resolved in a plaintiff's favor by a preponderance of the evidence, is not tantamount to proving guilt of perjury. See id. at 520, 113 S.Ct. at 2754. Second, he notes, a company might rely in good faith on a false statement of a front-line supervisor's professed belief in the reason given for some adverse employment action. See id. at 520-21, 113 S. Ct. at 2754-55. The supervisor, Justice Scalia maintains, may be a liar, but the company is not (though under agency principles the company may be liable for the supervisor's lie). In any event, Justice Scalia does not doubt that a proffered explanation, found by a fact-finder not to be the reason believed by the defendant, is a pretext, and (whether perjurious or not) supports an inference, when coupled with the facts of a prima facie case, that the true reason was discrimination.

The reason for permitting an inference of discrimination from a finding of pretext is evident. In the context of Title VII lawsuits, the likely motivation for a defendant, called upon in court to proffer an explanation for its adverse employment action, to proffer a pretextual explanation is to hide the true explanation of discrimination. At least, this is generally so in the absence of some plausible explanation for the pretextual proffer. As we have previously observed, "Resort to a pretextual explanation is, like flight from the scene of a crime, evidence indicating consciousness of guilt, which is, of course, evidence of illegal conduct." Binder v. Long Island Lighting Co., 57 F.3d 193, 200 (2d Cir. 1995). In McDonnell Douglas, 411 U.S. at 805, 93 S.Ct. at 1825-26, the Supreme Court called a pretextual proffer a "coverup" for discrimination. Of course, just as flight from the scene of a crime might be explained, there might be innocent explanations for proffering a reason not believed to be true, and evidence of such explanations will weigh against drawing the inference of discrimination.

Thus, the panel opinion is quite wrong to assert that the pretext finding made by the District Court in this case "points nowhere." On the contrary, it starts out pointing in the same direction that all pretext findings point--toward the finding of discrimination that is inferable from the facts constituting the plaintiff's prima facie case.8 That proposition has been consistently recognized by the Supreme Court and by every appellate judge that has considered a Title VII case. Whether the pretext finding continues to point toward discrimination with sufficient probative force to persuade a fact-finder to infer discrimination depends on the evidence in the case, but there can be no doubt of at least the initial direction in which a pretext finding points. When the Supreme Court says in St. Mary's, as all other courts have also said, that a fact-finder may infer discrimination from a finding of pretext, it is obviously in no doubt as to the direction in which a pretext finding points. Moreover, the Supreme Court's assertion that an inference of discrimination may be drawn from a finding of pretext, without any additional evidence required from the plaintiff, necessarily means that the pretext finding generally points toward discrimination with considerable force, enough force to enable the plaintiff to win, except in those rare cases, considered below, where a finding of discrimination, following a finding of pretext, may fairly be viewed as clearly erroneous because of special circumstances existing in the record.

I have no doubt that in some cases, a fact-finder would be fully entitled to consider the probative force of a pretext finding to be substantially diminished or even eliminated. This could occur for any of three general reasons. First, the fact-finder might decline to draw the inference of discrimination from a finding of pretext because evidence in the record points persuasively to the existence of a third motive (neither the proffered explanation nor discrimination) as the true explanation for the defendant's adverse employment action. For example, the evidence might show that the employer gave a false explanation for discharging an employee (chronic lateness) in order to spare the employee the embarrassment of the true reason (stealing company funds). Many courts have recognized that the falsity of a defendant's proffered reason might not support an ultimate finding of discrimination in the face of substantial evidence blunting the normally probative force of the finding of pretext. See Rhodes v. Guiberson Oil Tools, 75 F.3d 989, 994 (5th Cir.1996) (in banc) (example of one of several proffered reasons shown to be questionable); Binder, 57 F.3d at 200 (example of pretextual reason explained by desire to protect business secret or reputation of employee); Woods v. Friction Materials, Inc., 30 F.3d 255, 260-62 & n. 3 (1st Cir.1994) (example of pretextual reason for a discharge accomplished to prevent disclosure of employer's embezzlement); see also Isenbergh v. Knight-Ridder Newspaper Sales, Inc., 97 F.3d 436, 442-43 (11th Cir.1996) (example of employer proffering reason, lateness, that was falsely overstated). Obviously, if the evidence points toward a motive other than discrimination, a pretextual explanation offers slight, if any, basis for inferring discrimination.

Second, the evidence might persuade the fact-finder that some component of the plaintiff's prima facie case is not established. For example, the evidence might show that the plaintiff is in fact not a member of the protected class against which discrimination was alleged.

Third, the evidence might persuade the fact-finder that the defendant has such a strong tradition of not making discriminatory employment decisions as to make it unlikely that it acted on the basis of a discriminatory motive in the plaintiff's case. For example, the evidence might show that an employer has hired and promoted an unusually large number of members of the group allegedly discriminated against and has also disciplined supervisors for isolated instances of discrimination. In such circumstances, even if a proffered explanation is found to be pretextual, a fact-finder would have a sound basis for declining to infer that the true explanation was discrimination.

These examples do not exhaust the possible circumstances in which the fact-finder could properly accord a pretext finding little, if any, significance. The important point, however, is that in all such circumstances, the diminished significance of a pretext finding arises from evidence in the record. Evidence, not speculation about possibilities, will normally be the soundest basis for diminishing or even eliminating the probative force of a finding of pretext.

Beyond relying on such evidence, a fact-finder is also entitled to decline to draw an inference of discrimination from a finding of pretext simply because of the fact-finder's own view of the unlikelihood that the pretextual explanation was proffered as a cover for discrimination. When trial judges regularly instruct juries to rely on their "common sense" or their "experience" in determining whether an inference, including an inference of motivation, ought to be drawn,9 they are inviting these fact-finders to apply their own individual views about how likely it is that a person acted on the basis of an alleged motivation. Bench trial fact-finders, no less than jurors, are entitled to rely upon their common sense and experience in deciding whether to draw an inference, including the decision whether to infer discrimination from a finding of pretext.

Some might think that my view permitting the force of a pretext finding to be substantially diminished or even eliminated by the defendant's evidence of a "third motive" or any other circumstance that weakens the inference available from a pretext finding would be an instance of placing on an employer a burden to disprove discrimination. This is not so. To avoid entry of judgment as a matter of law, a defendant has no obligation to do anything other than proffer evidence of an explanation for its adverse employment action. However, if it proffers such evidence, it takes the risk that a fact-finder will find its explanation to be a pretext and the further risk that a fact-finder will infer discrimination from the facts of the prima facie case, strengthened by the permissible adverse inference from proffering a pretextual explanation. The employer is free to mitigate that risk by introducing evidence that explains why it proffered a pretextual explanation. It has no obligation or burden to do so, but, like all litigants facing the risk of liability based on permissible inferences from an adversary's evidence, it declines to present its own opposing evidence at its peril, even though the burden of persuasion on the ultimate issue remains on the plaintiff.

The point is best illustrated by the numerous cases permitting adverse inferences to be drawn against defendants even in criminal cases unless they provide an explanation that satisfies the jury. See, e.g., Turner, 396 U.S. at 405-18, 90 S.Ct. at 646-53 (possession of heroin may authorize conviction for importing heroin unless defendant satisfactorily explains possession); United States v. Gainey, 380 U.S. 63, 85 S. Ct. 754, 13 L. Ed. 2d 658 (1965) (presence at still being operated may authorize conviction for operating still unless presence satisfactorily explained by defend ant). Though two members of the Court believed that permitting such an inference in the absence of a defendant's satisfactory explanation shifted the burden of proof to the defendant, Turner, 396 U.S. at 432-35, 90 S.Ct. at 660-61 (Black, J., with whom Douglas, J., joins, dissenting); see also Gainey, 380 U.S. at 71, 85 S.Ct. at 759-60 (Douglas, J., dissenting in part on self-incrimination grounds); id. at 74, 85 S.Ct. at 761 (Black, J., dissenting on various grounds), the Supreme Court disagreed.

To say that a set of facts (subsidiary facts of the prima facie case plus a finding of pretext) permits an inference of discrimination unless the defendant presents evidence undermining the prima facie case or overwhelming it with contrary evidence does not assign to the defendant a burden of proof. The defendant has simply been afforded the opportunity of every defendant, confronted with a plaintiff's sufficient evidence, to present opposing evidence, without which it takes the risk that the trier might draw the inference of liability from the plaintiff's sufficient evidence. Neither the opportunity to present such evidence nor the risk that the inference might be drawn in its absence has ever been thought to place a burden of proof on the party opposing the inference.

For all of these reasons, I disagree profoundly with the majority's view that a finding of pretext in a discrimination case is often to be accorded little significance.10

(c) The Combined Effect of a Prima Facie Case and a Finding of Pretext. Having embraced a new and weak meaning of "prima facie case" and then depreciated the significance of a pretext finding simply because unproven possible explanations might exist, the majority then aggregates its views to assert that "the combined effect of both [a prima facie case and a finding of pretext] may have little capacity to prove what the plaintiff has the ultimate burden of proving." 114 F.3d at 1338. This view is completely at odds with Justice Scalia's statement in St. Mary's that "rejection of the defendant's proffered reasons will permit the trier of fact to infer the ultimate fact of intentional discrimination, and ... no additional proof of discrimination is required." St. Mary's, 509 U.S. at 511, 113 S.Ct. at 2749 (internal quotation marks, footnote, and brackets omitted; emphasis in original).

Though I agree with the majority that "a finding of pretext, together with evidence comprising a prima facie case, is not always sufficient to sustain an ultimate finding of intentional discrimination," 114 F.3d at 1343 (emphasis added), it will be a rare case where this is not so. Surely the three cases11 cited by the majority for the language just quoted are not examples of such insufficiency because in none of them was there any pretext finding at all, much less a pretext finding diminished by opposing evidence. As the Fifth Circuit has noted, "In tandem with a prima facie case, the evidence allowing rejection of the employer's proffered reasons will often, perhaps usually, permit a finding of discrimination without additional evidence." Rhodes, 75 F.3d at 994.

Whether this appeal presents one of those rare cases, explicitly contemplated by St. Mary's, 509 U.S. at 524, 113 S.Ct. at 2756, where a finding of discrimination may be deemed clearly erroneous despite a valid finding of pretext, requires consideration of how an appellate court reviews findings of fact, including a finding of discrimination.

On the aspect of the appeal that concerns appellate review of a trial court's findings of fact, we again start in complete agreement. We all agree that any finding of fact by a district judge may be rejected on appeal if the finding is "clearly erroneous," see Fed.R.Civ.P. 52(a), a conclusion that may be reached when the appellate court is "left with the 'definite and firm conviction that a mistake has been committed.' "12 See Inwood Laboratories, Inc. v. Ives Laboratories, Inc., 456 U.S. 844, 855, 102 S. Ct. 2182, 2189, 72 L. Ed. 2d 606 (1982) (quoting United States v. United States Gypsum Co., 333 U.S. 364, 395, 68 S. Ct. 525, 542, 92 L. Ed. 746 (1948)). Furthermore, we all agree that this basic principle of appellate review applies to findings of fact about a defendant's state of mind in discrimination cases. The Supreme Court settled that point fifteen years ago. See Pullman-Standard v. Swint, 456 U.S. 273, 287-88, 102 S. Ct. 1781, 1789-90, 72 L. Ed. 2d 66 (1982) (intent to discriminate is issue of fact, subject to "clearly erroneous" standard of review). Finally, we all agree that clear error review is available for a finding of discrimination made after a finding of pretext. As the Supreme Court stated in St. Mary's, even after a finding of pretext, the plaintiff's claim of discrimination "remains a question for the factfinder to answer, subject, of course, to appellate review--which should be conducted ... under the 'clearly erroneous' standard of Federal Rule of Civil Procedure 52(a)." St. Mary's, 509 U.S. at 524, 113 S.Ct. at 2756 (citing Anderson v. City of Bessemer City, 470 U.S. 564, 573-76, 105 S. Ct. 1504, 1511-13, 84 L. Ed. 2d 518 (1985)).

We divide, however, not on whether an appellate court may review a finding of discrimination to determine if it is clearly erroneous, but on how such review is to be made. An appellate court may not simply review a record and announce that it is left with a firm conviction that a mistake has been made. After a district judge has made an ultimate finding of fact and has supported that finding with a large number of subsidiary findings of fact, proper regard for the relationship between a court of appeals and a district court requires more than a bare assertion of disagreement. The reviewing court has an obligation either to analyze the evidence relied on to support the ultimate finding and explain why it is seriously deficient, or to analyze other evidence in the record and explain why it so substantially supports a finding contrary to the one reached by the district judge as to create a firm conviction that the judge's finding is wrong. See Anderson, 470 U.S. at 575, 105 S.Ct. at 1512. In rare cases, both tasks might lead to appellate rejection of a finding.

Appellate courts rejecting district court findings as clearly erroneous have regularly undertaken these tasks. See, e.g., Dayton Board of Education v. Brinkman, 443 U.S. 526, 534-37, 99 S. Ct. 2971, 2977-79, 61 L. Ed. 2d 720 (1979) (approving Court of Appeals rejection of finding after identifying significant evidence overlooked by District Court); Jiminez v. Mary Washington College, 57 F.3d 369, 380-82 (4th Cir.1995) (rejecting finding of discrimination after analyzing and deeming insubstantial evidence trial court relied on and also analyzing substantial contrary evidence); Sumner v. United States Postal Service, 899 F.2d 203, 209-11 (2d Cir.1990) (analyzing evidence opposing District Court's rejected finding that employer's proffered reason for discharge was not pretext).

In this case, the panel decision that the in banc majority has permitted to go into effect did not perform either task. It did not even purport to deem deficient the evidence underlying either the plaintiff's prima facie case of discrimination against married women or the finding of pretext. Not one of the District Court's 111 subsidiary findings of fact was ruled unsupported by the evidence. Nor did the panel attempt to marshall evidence in the record opposed to the ultimate finding of discrimination.

This does not mean that a fact-finder's inference of discrimination is immune from review for clear error. An appellate court may reject an inferential ultimate finding if the subsidiary facts are clearly erroneous, if there is a complete absence of subsidiary facts, e.g., Goldhirsh Group, Inc. v. Alpert, 107 F.3d 105, 109-10 (2d Cir.1997) (inference of content of telephone conversation rejected for lack of factual predicate), or if the fact-finder articulated, or necessarily relied on, an aspect of its reasoning process that was not merely different from the views of the appellate court but was demonstrably unsound. But the rejection must be based on evidence in the record.

Our disagreement on the circumstances under which an inference of discrimination may be rejected is at the heart of the matter. The majority believes that an appellate court may reject a fact-finder's inference of discrimination, drawn from a finding of pretext, whenever the appellate court thinks (or in Rule 52(a) terms, "is firmly convinced") that reasons other than discrimination are as likely to explain a pretextual proffer as discrimination itself. No authority is cited for this proposition, not any Title VII case, not even any decision from any other field of substantive law. I fundamentally disagree with the majority's unprecedented grant to reviewing panels of a roving commission to second-guess district courts on the relative likelihood of possible explanations for a Title VII defendant's proffer of a pretextual reason. A reviewing court may legitimately reject an inference of discrimination, not because of its views about employer behavior in general or university behavior in particular, but only where a record provides an articulable basis for exercising the narrowly circumscribed appellate function of determining that a finding of fact is clearly erroneous.

It is not difficult to imagine cases where, despite a sustainable finding of pretext, a record contains evidence that justifies an appellate court in ruling that an ultimate finding of discrimination is clearly erroneous. Just as the fact-finder is entitled to decline to draw an inference of discrimination from a finding of pretext, an appellate court is entitled to conclude from evidence in the record that the inference of discrimination is so demonstrably unsupportable as to be clearly erroneous. Such clear error could occur if the record overwhelmingly established any of the circumstances discussed above that would entitle a fact-finder not to infer discrimination from a pretextual proffer. These include facts that overwhelmingly establish a third motive for the pretextual proffer, undermine a component of the plaintiff's prima facie case, or demonstrate that the defendant is highly unlikely to have discriminated. But in the absence of such evidence, a reviewing court may not reject a finding of discrimination simply because the reviewing court does not believe that the inference of discrimination is warranted. That is the clear message of Anderson, where the Supreme Court reversed a court of appeals for disagreeing with a district court's inference of discrimination, which was based on pretextual explanations. "This standard [a definite and firm conviction that a mistake has been made] plainly does not entitle a reviewing court to reverse a finding of the trier of fact simply because it is convinced that it would have decided the case differently." Anderson, 470 U.S. at 573, 105 S.Ct. at 1511.

The point is well illustrated by the majority's comment on the significance of an inference of guilt arising from flight from the scene of a burning building: "[F]light from the scene of a crime ordinarily has evidentiary weight, but flight from a scene of arson shows nothing if the defendant fled the scene of a department store inferno during business hours." 114 F.3d at 1346. What diminishes the significance of flight in this example is evidence that establishes key facts: (a) the building is a department store, and (b) the suspect fled from it during business hours. That evidence, not an appellate court's generalized view of why people flee from crimes (or proffer pretextual explanations for tenure decisions) diminishes the significance of daytime flight from a burning department store, and even permits an appellate court to deem an inference of guilt from such flight clearly erroneous.

Having disclaimed any review of the adequacy of the District Court's finding of discrimination,13 the majority instead has elected to narrow the scope of the in banc appeal and authorize issuance of a mandate that relies on the rulings of the panel that initially heard this appeal. That procedural device requires consideration of the third matter presented by this case--an in banc court's relation to a panel opinion.

The majority observes, and I fully agree, that an in banc court is not normally convened to reconsider a panel's assessment of a district court's factual findings. In this case, however, an in banc court was convened in an appeal that largely concerned the panel's rejection of the District Court's finding of discrimination. The rehearing order contained no limitation of issues. See Fisher v. Vassar College, No. 94-7737 (2d Cir. Feb. 16, 1996) (order for rehearing). Once an appeal is ordered for a plenary rehearing, the in banc court normally ought to make its independent assessment of the appeal.14 Of course, an in banc court has the option of deciding only certain issues and then voting to dissolve itself and return any undecided issues to the panel for further consideration in light of the decision of the in banc court, cf. Peck v. United States, 102 F.3d 1319 (2d Cir.1996) (in banc) (in banc court dissolves itself and returns appeal to panel for reconsideration in light of recent Supreme Court decision), but the in banc court has not done so in this case.

Instead, the majority pursues a most unusual course. Despite the grant of plenary in banc review, the majority first frames a limited issue for in banc decision, then states that the majority's "ruling" on the limited issue "reaffirms the applicability of the rule employed by the panel," 114 F.3d at 1347, and finally orders issuance of a mandate that draws its authority both from the in banc opinion and the rulings of the panel as originally made, without any further consideration.

The majority's unprecedented approach is not even internally consistent. At the outset, the majority states, "A majority of the Court has decided to limit in banc review to resolution of the question whether a finding of liability under Title VII, supported by a prima facie case and a sustainable finding of pretext, is subject to review for clear error." 114 F.3d at 1333 (footnote omitted). Then, after framing this narrow (and indisputable) question, the majority moves on to consider several other matters. The majority states that a prima facie case of discrimination, considered apart from the persuasive force of opposing evidence, is not necessarily sufficient to take a case to a fact-finder. Then, acknowledging that the in banc rehearing is not limited to the question of whether clear error review is possible, as it first said, the majority states that the review is "limited to the force and effect of a pretext finding." Id. at 1347. On this issue, the majority further states that a finding of pretext has little, if any, significance whenever an appellate court believes that there are possible explanations, even though not established in the evidence, that might explain why an employer proffered a false explanation for its action. Later, the majority, though disclaiming the promulgation of any "rule," id. at 1346 ("The rule is that there is no rule peculiar to discrimination cases."), announces a novel method of analyzing an employer's false statement, which looks very much like a "rule": "the weaker the evidence of discrimination, the less reason there is to believe that the employer's false statement concealed discrimination," id. at 1346-47. Finally, the majority, though declining to consider the correctness of the panel's rejection of the District Court's finding of discrimination, undertakes to explain what the panel must have meant by its extraordinary statement that a finding of pretext "points nowhere." Id. at 1345.

The majority's novel method of proceeding raises two serious issues of institutional concern: (a) what issues should this in banc court adjudicate, and (b) what procedure should be required of the panel after dissolution of the in banc court?

(a) Issues for the in banc court. The majority asserts that the in banc rehearing was ordered to resolve the issue of "whether a finding of liability under Title VII, supported by a prima facie case and a sustainable finding of pretext, is subject to review for clear error." 114 F.3d at 1333. That is assuredly not why I voted for in banc review, since that issue had previously been decisively settled by St. Mary's. See 509 U.S. at 524, 113 S. Ct. at 2756. My vote to in banc, and, I believe, the votes of some other members of the Court, were cast to determine not whether clear error review was available, but how such review should be conducted. Many of us were deeply troubled by the panel's unprecedented ruling that a finding of pretext "points nowhere," see Fisher II, 70 F.3d at 1437, which was contrary to the ruling in St. Mary's that such a finding permits an ultimate finding of discrimination.

Even if many of those voting to in banc this appeal did so primarily, or even only, to confirm the proposition that clear error review is permissible, once the appeal is reheard, we should discharge our responsibilities by facing up to major issues fairly presented by the case. Manifestly, a major issue arising from the panel opinion is what significance a reviewing court should give to a sustainable finding of pretext. Does it "point[ ] nowhere," as the panel stated? Does it point toward discrimination, but with only minimal probative force, as perhaps some believe? Or does it point toward discrimination with strong probative force, at least in the absence of evidence that blunts such probative force, as I believe? Rather than answer these questions, the majority expresses a few thoughts about a pretext finding and then declines to rule on whether or not the District Court's use of a pretext finding to support an inference of discrimination in this case is permissible. But whether such use is permissible is the crucial issue on this appeal. To decline to rule on it is a default of adjudication. The district judges of this Circuit gain little, if any, illumination from learning that a majority of this Court believes that lack of candor might explain why a Title VII defendant proffers a reason that is pretextual. What they need to know, especially after reading the panel opinion in this case, is whether this Circuit holds the view that a pretext finding points nowhere, and, if not, to have some guidance as to the appropriate role of such a finding in the resolution of a Title VII claim. This in banc rehearing was the occasion for providing that guidance. Regrettably, the opportunity has been missed.

Rather than provide such guidance, the in banc court has voted to dissolve itself without considering whether the pretext finding in this case, plus other findings of the District Court, support the ultimate finding of discrimination. Though the majority has the undoubted power to use its one-vote edge to accomplish this result, its action is an inappropriate use of the in banc procedure. A clear majority of the active judges voted to accord this appeal a plenary rehearing. No changed circumstance has arisen since that decision to warrant a restriction of the scope of the in banc.15 The majority has simply elected to use its voting power to remove from in banc consideration the resolution of the issue that occasioned the in banc order.16 In the Supreme Court, a minority of four justices are permitted to bring a case to the Court for review, after which the majority adjudicates the issues presented.17 See Leonard M. Leiman, The Rule of Four, 57 Colum. L. Rev. 975 (1957). In this case, despite a majority vote of the active judges in favor of in banc review, a narrow majority of the in banc court has now voted to remove the basic issue on this appeal from consideration by the in banc court.

I do not contend that all members of an appellate court, and certainly not of an in banc court, are obliged to read every word of a record in order to cast a responsible vote on the issue of whether an ultimate finding of fact is clearly erroneous. Judges are entitled to rely on the materials culled from the record by knowledgeable appellate lawyers on opposing sides. They are also entitled to accept, without detailed independent verification, an analysis of the evidence set forth in majority and minority opinions. But to decline to use any of these traditional techniques and instead to remove the key issue of this appeal from in banc consideration is unfair both to the judges of this Court who wish to have that issue considered (whether or not their views on the merits ultimately prevail) and to the judges of the district courts who will now not receive the authoritative guidance for which this in banc rehearing was ordered.18

(b) Procedure for the panel after in banc dissolution. After limiting the scope of the in banc rehearing and dissolving the in banc court, the majority then takes the wholly unprecedented step of ordering a mandate to issue without the need for reconsideration by the panel in light of the in banc majority's opinion. This step is taken because of the in banc majority's assertion that its "ruling" on the now limited question of "the force and effect of a pretext finding ... reaffirms the applicability of the rule employed by the panel." 114 F.3d at 1347.

In fact, it is evident that, at least in two crucial respects, the panel did not proceed as the majority now indicates is required of a reviewing court diminishing the significance of a pretext finding. First, the in banc majority opinion says that a finding of pretext may be accorded little, if any, weight where circumstances make it clear that the probability that the pretext was proffered to hide discrimination is just as likely as several other explanations that the defendant might have had for such a proffer. But in this case, the fact-finder made no such finding of equal probabilities, and the panel that rejected Judge Motley's finding of discrimination made no determination that it thought that the probabilities were equal. Second, the in banc majority says that appellate rejection of a finding of discrimination must be based on the evidence in the record. But in this case the panel opinion did not point to evidence that undermined the plaintiff's prima facie case, supported a non-discriminatory reason for proffering a pretextual explanation, or showed that discrimination was not the reason for the tenure denial. Of course, it is possible that if this appeal is returned to the panel for reconsideration in light of the majority's in banc opinion, the panel would again conclude that the finding of discrimination was clearly erroneous. But adjudication of this appeal should not rest on the majority's prediction of what the panel will do.

The reality is that the issue framed by the majority--"the force and effect of a pretext finding"--is the type of issue that gains only slight illumination from abstract statements. Only the exercise of an appellate court's authority to review a discrimination finding for clear error will oblige that court to explain how such review is being conducted and why, in a particular case, a pretext finding lacks the "force and effect" that normally permits it to support a discrimination finding. Since the majority has said so little about how a discrimination finding is to be reviewed and how a finding of pretext may be diminished in significance, it is not surprising that it prefers not to demonstrate the shortcomings of its "ruling" by attempting to apply it to Judge Motley's findings of pretext and discrimination.

Since the majority authorizes issuance of a mandate that draws part of its authority from the panel opinion, I am obliged to turn my attention to the reasons given by the panel for rejecting the District Court's finding of discrimination.

4. The Panel Opinion's Rejection of the Finding of Discrimination19

The panel opinion starts with the District Court's finding of pretext and then abruptly eliminates it from the ensuing analysis by stating, "The finding of pretext affirmed in this opinion points nowhere...." Fisher II, 70 F.3d at 1437. As discussed above, this total rejection of the significance of the pretext finding cannot be reconciled with the contrary view of the Supreme Court in St. Mary's.

What then was there about the pending case that led the panel to ascribe no weight to the finding of pretext and to rule that the finding of discrimination was clearly erroneous? The panel took two further steps after disregarding the finding of pretext. First, it said that the District Judge herself did not believe that the combination of a prima facie case plus a finding of pretext supported an ultimate finding of discrimination. Id. I see nothing in Judge Motley's opinion to support this statement, and, even if some basis for doubt on this score existed, the proper remedy would not be to reverse the District Judge by reading (or misreading) her mind, but to remand the case so that Judge Motley can tell us whether she finds that the prima facie case plus the finding of pretext suffices to persuade her, as the fact-finder, to make the ultimate finding of discrimination.

Lacking any words in Judge Motley's opinion to support the notion that she did not believe the pretext finding supported an inference of discrimination, the panel bases its view of Judge Motley's thinking on the fact that after finding both a prima facie case and pretext, she went on to identify three categories of evidence that lent additional support to her ultimate finding of discrimination. As the panel views it, Judge Motley "felt compelled to comb the record for evidence of discrimination, evidently convinced, as are we, that the finding of pretext here did not alone justify a finding of discrimination." Id. This is an unfair characterization of Judge Motley's opinion. The District Judge proceeded as bench trial judges frequently do: she made the two findings that the case law of the Supreme Court and this Circuit says are sufficient to support an inference of discrimination, she then drew the permissible inference from pretext and made the ultimate finding of discrimination, and then she went on to make additional findings to demonstrate added support for her ruling.

The panel's next step was to reject, not the basic findings of a prima facie case and pretext, but some of the additional findings that lend further support to the ultimate finding of discrimination. Judge Motley discussed three categories of evidence as additional support for her finding of discrimination--(a) discriminatory use of student evaluations, (b) statistical evidence, and (c) anecdotal evidence. Fisher v. Vassar College, 852 F. Supp. 1193, 1228-29 (S.D.N.Y.1994) ("Fisher I "). The panel rejected the statistical evidence, Fisher II, 70 F.3d at 1442-47, and the anecdotal evidence, id. at 1438-40, and, for purposes of this discussion, I will assume, only for the sake of the argument, that the District Court's findings in these two categories were clearly erroneous. But the panel disregarded, without explanation, Judge Motley's first category of additional evidence. This concerned student evaluations, which she found were "almost exclusively" the basis for assessment of Dr. Fisher's teaching performance. Fisher I, 852 F. Supp. at 1209 (finding 45).

Judge Motley found that Vassar students rated teachers for various characteristics on a scale of one to five, and that for male tenure candidates and non-married female tenure candidates Vassar considered the percentage of students who rated candidates in the top two categories but that for Dr. Fisher, the only married female considered for tenure, Vassar considered only the percentage of students rating her in the one highest category. As a result, Dr. Fisher had a lower percentage of favorable student evaluations than males and unmarried women, and this lower percentage was considered in the tenure evaluation.20 Id. at 1209-10 (findings 47-49), 1228. This additional finding was not ruled clearly erroneous by the panel, and no explanation was given as to why it does not lend some weight to the finding of discrimination. Indeed, this subsidiary finding was accepted by the panel in its discussion of the evidentiary sufficiency of the finding of pretext, Fisher II, 70 F.3d at 1435-36, but omitted, without explanation, in the panel's rejection of the District Judge's finding of discrimination.21

Though leaving undisturbed the findings about discriminatory use of student evaluations, the panel then proceeded to reject aspects of Judge Motley's findings that she had not included in the three categories of evidence that she elected to discuss as giving added weight to her finding of discrimination. These other aspects concern expert testimony, id. at 1447-48, and party admissions, id. at 1440-42. Again, for purposes of this discussion, I will assume that the panel was entitled to reject these other findings. Nevertheless, the basic issue remains: how can a panel reject a finding of discrimination that is based on the facts that establish a prima facie case, plus a finding of pretext, plus the undisturbed additional finding of discriminatory use of student evaluation ratings?

The panel's approach boils down to three steps: (a) the finding of pretext is disregarded, (b) the findings of discriminatory use of student evaluations are ignored, and (c) other findings are found to be clearly erroneous. This approach is inconsistent with every instance of appellate review of trial court fact-finding in discrimination cases. Even if some of the trial court's subsidiary findings are deemed clearly erroneous, the most an appellate court could properly do would be to remand to the District Court to reconsider its ultimate finding of discrimination on the basis of the subsidiary findings that are undisturbed--the facts of the prima facie case, the finding of pretext, and the findings of discriminatory use of student ratings. Since Judge Motley's opinion leaves me in no doubt that, even if limited to these undisturbed matters, she would infer discrimination against married women, I would affirm this aspect of the judgment she entered.

Moreover, there is an additional basis in the evidence and the findings that supports Judge Motley's ultimate finding of discrimination against married women. Judge Motley found evidence of bias against married women reflected in the adverse comments in the tenure review file concerning Dr. Fisher's eight-year hiatus from teaching in order to raise a family. Fisher I, 852 F. Supp. at 1216 (findings 84-87). This hiatus occurred from 1966 to 1973, after which Dr. Fisher resumed teaching, first at Marist College for three years and then at Vassar. Though fully aware that Dr. Fisher had been away from teaching for an eight-year period that ended three years before she applied to Vassar, Vassar hired her and placed her on a tenure track. As Judge Motley noted:

The persistent fixation of the Biology Department's senior faculty on a married woman's pre-Vassar family choices reflects the acceptance of a stereotype and bias: that a married woman with an active and on-going family life cannot be a productive scientist and, therefore, is not one despite much evidence to the contrary.

Id. (finding 87).

Though the panel opinion discussed the evidence that Judge Motley relied on to support her findings regarding Dr. Fisher's hiatus from teaching and acknowledged that "[t]he evidence supports an inference that Fisher's eight-year absence from academia hurt her chance for tenure," 70 F.3d at 1448, it concluded that the District Court's inferences about discrimination from such evidence were not supportable. The panel expressed the view that the adverse comments about her hiatus were not "sex specific" and were legally insufficient to lend support to a Title VII claim. Fisher II, 70 F.3d at 1448. The panel agreed with Vassar's contention that the hiatus would be relevant to her claim only if the tenure experiences of women who took extended leaves of absence compared unfavorably with the tenure experiences of men who took such absences. Id.

I strongly disagree with this assessment of evidence that Judge Motley properly deemed significant. The panel's approach means that working women, penalized for taking years off to raise children, can be disadvantaged in the workplace for so long as it takes our society to become accustomed to the idea that working men share a responsibility to devote their time to raising children. Of course, Vassar was careful to say that the relevant comparison, necessary to make the adverse comments probative, should be between men and women who took extended absences from work regardless of the reason. Such a felicitous generalization cannot mask the obvious truth that the predominant reason that working women take absences from work is to bear and raise children; there is no pervasively comparable reason for absences among working men, and it is fatuous for the college to suggest that Dr. Fisher is to be faulted for not producing evidence of the tenure experiences of men who took absences "regardless of the reason." With rare exception, men do not take extended absences from work to raise children (or for any other reason); their reluctance to share in child-raising responsibilities cannot fairly be turned against Dr. Fisher to penalize her for not assembling data concerning such non-existent male absences. Judge Motley was entirely correct: the Biology Department's "persistent fixation on [Dr. Fisher's] pre-Vassar family choices reflects the acceptance of a stereotype and bias."

Of course, a college, or any other employer, is entitled to give an unfavorable rating to a person whose absence from a field, for whatever reason, has left that person deficient in necessary knowledge of current learning and developments. But there was not even a claim, much less evidence, that Dr. Fisher's child-raising hiatus had left her in any respect deficient in contemporary knowledge of her field. On the contrary, the record is replete with evidence of her numerous peer-reviewed publications, grant awards, papers presented, and consultantships to such institutions as the National Institutes of Health and the National Science Foundation. Fisher I, 852 F. Supp. at 1198-1205 (findings 9-31). Judge Motley was entirely justified in considering the adverse view of Dr. Fisher's hiatus to be further evidence of bias against married women.22

The District Court also found that Vassar had denied Dr. Fisher tenure on the basis of age. Fisher I, 852 F. Supp. at 1230-31. This second ultimate finding was based primarily on (a) a prima facie case of age discrimination and (b) a finding that Vassar's proffered reason for denying tenure--failure to meet tenure standards--was pretextual.23 Id. Significantly, the District Court found that a component of the prima facie case of age discrimination--discharge under circumstances giving rise to an inference of age discrimination--was established by evidence showing that, with one distinguishable exception, see id. at 1230 n. 24, "all other tenured faculty who were equally or less qualified than Dr. Fisher were at least nine years younger than Dr. Fisher when they were tenured." Id. at 1230; see id. at 1219 (finding 106). Moreover, had Dr. Fisher not taken eight years off from teaching to raise her children, before joining the Vassar faculty, she would have been within the age group from which Vassar is apparently willing to select tenured professors.

The claim of discrimination on two distinct grounds, inferable from a finding of pretext, introduces an added complexity to this case. Though the analogy is not exact, the situation is somewhat comparable to that presented by a person who is apprehended while fleeing from the scene of two crimes. The fact-finder may draw from flight an inference of consciousness of guilt with respect to one of the crimes, the other crime, or both crimes. However, it is also true that, in some circumstances, the strength of the inference from flight that tends to prove guilt of one offense makes it most unlikely that an inference of guilt as to the other crime is reasonable.

For example, if a person is apprehended fleeing from a scene where two victims have been robbed moments before, it is entirely reasonable to infer consciousness of guilt of both crimes, at least in the absence of circumstances significantly distinguishing the offenses. On the other hand, if the person fled from a scene where a bank was robbed and a work of art was elaborately defaced, and if circumstances strengthened the inference of the person's guilt of the bank robbery, the very act of fleeing from the bank robbery would make it highly unlikely that the fleeing person paused to commit an act of vandalism. In such circumstances, the trier would be fully justified in drawing an inference of guilt only as to the robbery, and, depending on the facts, an additional inference as to the vandalism, if drawn by the fact-finder, might well be clearly erroneous.

In the pending case, where the plaintiff presented a prima facie case of discrimination because she was a married woman and because of age, it was within the prerogative of the fact-finder to draw from the finding of pretext an inference of discrimination on both grounds. Unlike the bank robbery/vandalism example discussed above, discrimination based on marital status is not realistically incompatible with additional discrimination based on age. An employer could rationally decide to deny tenure to married women and to those over 40, and would therefore likely deny tenure to a married woman over 40. Thus, this is not a case where the very existence of one discriminatory motive is itself strong evidence that another alleged discriminatory motive is unlikely to have existed. Though the evidence more strongly supported the ultimate conclusion of discrimination against married women, it also permitted the inference of discrimination on the basis of age.

The majority has permitted the rejection, as clearly erroneous, of findings of discrimination that are entirely permissible in view of the fact that the finding of pretext is undisturbed by the Court, unexplained by the defendant, and uncontroverted by evidence in the record. That rejection is unwarranted, and the majority offers no satisfactory explanation for what it has done. In the absence of such explanation, one can only speculate about what underlies the majority's decision. The most likely explanation is that the majority is reluctant to have a court confer tenure on a professor who has been rejected for tenure by her faculty colleagues, her dean, her college president, and her board of trustees.24 I share that reluctance, and had I been the trier of fact, I might have declined to find pretextual the proffered explanation of noncompliance with tenure standards, or to draw the inference of discrimination, even after a finding of pretext. But, as we have been authoritatively instructed, "the court of appeals may not reverse [a finding] even though convinced that had it been sitting as the trier of fact, it would have weighed the evidence differently." Anderson, 470 U.S. at 574, 105 S.Ct. at 1511.

Though it is also possible that the majority prefers to fashion a less deferential approach to appellate review of a discrimination finding because the adverse employment decision has been made by a highly respected college, or because the particular action taken is the denial of tenure, we have no authority to apply special rules of appellate review for such reasons.25 See University of Pennsylvania v. EEOC, 493 U.S. 182, 190, 110 S. Ct. 577, 583, 107 L. Ed. 2d 571 (1990) ("The effect of the elimination of this exemption [for educational institutions] was to expose tenure determinations to the same enforcement procedures applicable to other employment decisions.") (emphasis added); Zahorik v. Cornell University, 729 F.2d 85, 93 (2d Cir.1984) ("Tenure decisions are not exempt under Title VII...."); Powell v. Syracuse University, 580 F.2d 1150, 1154 (2d Cir.1978) ("[A]cademic freedom [does not] embrace[ ] the freedom to discriminate.").

In the end, I am left with no adequate basis to reject the findings of married-woman and age discrimination made by the conscientious District Judge after a meticulous analysis of an extensive record. Even if, as I suspect, the majority's willingness to tolerate the rejection of these findings is an aberrational instance of excessive appellate review, prompted by reluctance to upset a college's denial of tenure, I cannot agree that such a rejection comports with applicable principles of law, and I therefore respectfully dissent.

WINTER, Circuit Judge, with whom Chief Judge JON O. NEWMAN, Judges KEARSE and CABRANES concur, dissenting:

Although I concur in Chief Judge Newman's dissenting opinion, I write separately because I hold views that are not fully reflected in that opinion.

The in banc court is unanimous in agreeing that Fisher established "a prima facie case," in some sense of the term. The in banc court is also unanimous in affirming as not clearly erroneous Judge Motley's finding that Vassar's proffer of an explanation was not the "real reason" but "a pretext" for denying Fisher tenure. Agreement ceases at this point. The majority takes the view that a Title VII/ADEA plaintiff who has: (1) established a prima facie case, (2) shifted the burden of production to the employer, (3) demonstrated an unrehabilitated pretextual explanation by that employer, and (4) obtained a finding from the trier of fact that the employer acted out of an unlawfully discriminatory motive, may still lose on the ground that the evidence of discrimination was legally insufficient.

Much of my disagreement with my colleagues in the majority stems from their narrow focus on selective quotations from Supreme Court Title VII cases. I do not deny that they find legitimate comfort in those quotations. However, the very same cases contain other language that is equally or more supportive of the opposite conclusion. It is hardly necessary, or even appropriate, however, to rely exclusively upon cut-and-paste analysis of Title VII decisions to address the legal issues that divide us. Those issues are not peculiar to discrimination law; rather, they involve the law of evidence and arise in every area of substantive law. Concepts such as prima facie case, presumptions, burden of production, and pretext have been the subject of a vast number of judicial decisions and academic discussions. My colleagues in the majority avoid a discussion of the implications from other areas of the law for the use of these concepts in Title VII/ADEA cases simply by declaring that these concepts are "quite different" in discrimination law. Finding little in Supreme Court decisions that supports that conclusion, I believe that the use of these concepts in the other areas of the law must be addressed.

"Prima facie case" has two meanings, one stronger than the other: (i) the plaintiff has produced evidence sufficient to establish a disputed fact from which, if credited by the trier, arises a rebuttable presumption--the strong version; or (ii) the plaintiff has produced evidence sufficient to permit, but not compel, a trier of fact to find a disputed fact for the plaintiff (or party bearing the burden of persuasion on the disputed fact)--the weak version. See Texas Dep't of Community Affairs v. Burdine, 450 U.S. 248, 254 n. 7, 101 S. Ct. 1089, 1094 n. 7, 67 L. Ed. 2d 207 (1981); Christopher B. Mueller & Laird C. Kirkpatrick, Evidence, § 3.4, at 137-38 (1995).

Meaning (ii) is subsumed within (i). As the Supreme Court said in Burdine:

The burden of establishing a prima facie case of disparate treatment is not onerous. The plaintiff must prove by a preponderance of the evidence that she applied for an available position for which she was qualified, but was rejected under circumstances which give rise to an inference of unlawful discrimination.

450 U.S. at 253, 101 S. Ct. at 1094.

Proof of "circumstances which give rise to an inference of unlawful discrimination" is surely sufficient to get to a jury. Moreover, footnote 6 to the quoted passage makes it clear that the four factor test as illustrated in McDonnell Douglas Corp. v. Green, 411 U.S. 792, 93 S. Ct. 1817, 36 L. Ed. 2d 668 (1973)--the Burdine plaintiff was (1) a female (2) qualified for the job in question but (3) rejected (4) in favor of a male--is alone proof of "circumstances which give rise to an inference of unlawful discrimination." Burdine, 450 U.S. at 253 n. 6, 101 S. Ct. at 1094 n. 6. At the very least, therefore, "prima facie case" means that a party bearing the burden of persuasion has proffered legally sufficient evidence to permit a trier to find the disputed fact--here, unlawful discrimination. Mueller & Kirkpatrick, supra, §§ 3.2, 3.4.

Under McDonnell Douglas, a prima facie case is initially of the stronger (i) variety. Burdine, 450 U.S. at 254 n. 7, 101 S. Ct. at 1094 n. 7. If the employer remains silent and the trier finds the facts that trigger the McDonnell Douglas presumption, the employer loses. Id. at 254, 101 S. Ct. at 1094. The employer thus bears the burden of production regarding the reasons for an adverse employment decision. Id. If the employer proffers evidence of lawful reasons, we move to the next step of the analysis.

Title VII and the ADEA do not require an employer to proffer a reason for an adverse employment decision with which the trier of fact agrees. An employer is free to make mistakes. For example, a faculty may regard a tenure candidate's scholarship to be inadequate while the trier of fact believes that her writing is a major scholarly contribution. So long as the faculty truly, even if mistakenly, holds that negative view, there is no Title VII violation.

Once the employer proffers a lawful reason for the employment decision, then the "presumption [of unlawful discrimination] ... drops out." St. Mary's Honor Ctr. v. Hicks, 509 U.S. 502, 510-11, 113 S. Ct. 2742, 2749, 125 L. Ed. 2d 407 (1993) (citing Burdine, 450 U.S. at 255, 101 S.Ct. at 1094-95). The majority opinion holds that when the "presumption ... drops out," proof of the McDonnell Douglas four factors without more no longer constitutes a prima facie case of discrimination. The basis offered by my colleagues for this conclusion, apart from the "drops out" language itself, is the fact that the burden of persuasion remains with the Title VII/ADEA plaintiff after the employer has met its burden of production.

The issue is hardly disposed of so easily. The role of rebuttable presumptions involves a debate that has spawned a vast body of commentary and resulted in competing schools of thought traced to such venerable scholars as James Bradley Thayer and Edmund Morgan. See, e.g., Edmund M. Morgan, Basic Problems of Evidence 28-45 (1962); James B. Thayer, Preliminary Treatise on Evidence 346 (1898); Edmund M. Morgan, Instructing the Jury upon Presumptions and Burden of Proof, 47 Harv. L.Rev. 59, 82 (1933); see generally Charles McCormick et al., McCormick on Evidence § 344 (4th ed.1992); 1 Jack B. Weinstein et al., Weinstein's Evidence p 300 (1996). Variations abound within each school of thought, see, e.g., Mueller & Kirkpatrick, supra, § 3.8, at 148-50; McCormick et al., supra, § 344; Weinstein et al., supra, p 300, at 300-3 to 300-4, and there is no one solution of commanding persuasiveness, much less one to be found in Supreme Court Title VII decisions that contain language useful to all the antagonists.

In my view, the "drops out" language indicates only that the McDonnell Douglas rebuttable presumption, version (i), loses all evidentiary weight, leaving a prima facie case in the sense of version (ii). I reach this conclusion for the following reasons. "Prima facie case" and "presumption" are not synonymous, as Burdine footnote 7 indicates. 450 U.S. at 254 n. 7, 101 S. Ct. at 1094 n. 7. The term prima facie case, even when used in sense (ii), means that a party having the burden of persuasion regarding a disputed fact has presented legally sufficient evidence to allow the trier of fact to find that fact. Id.

Presumptions, on the other hand, are legal rules calling upon the trier to add weight to one party's evidentiary scale if the evidence of the basic facts triggering the presumption is sufficient to allow the trier to find those facts. See, e.g., Mueller & Kirkpatrick, supra, § 3.4; 1 Weinstein et al., supra, p 300. Most presumptions, however, are rebuttable, and there is no general agreement among commentators or courts at large what, if any, residual effect a rebutted presumption has. To say that a "presumption ... drops out," therefore, does not inexorably mean that there is no longer legally sufficient evidence--i.e., a prima facie case--to allow a trier to find the disputed fact; it may mean only that the trier now resolves the issue based on its view of the evidence as a whole without giving the facts constituting the prima facie case the added evidentiary weight that they formerly had.

The majority give two principal reasons why it believes that a Title VII/ADEA plaintiff's prima facie case disappears once the employer proffers evidence of any explanation for the adverse employment decision.

The first reason given is the familiar rule that the burden of persuasion remains with the Title VII/ADEA plaintiff even after the employer meets its burden of production. However, it is a non-sequitur to reason that, because the plaintiff bears the burden of persuasion once the defendant's production is satisfied, a bare McDonnell Douglas case is no longer legally sufficient; there is no inconsistency between the proposition that a plaintiff's prima facie case survives a defendant's satisfaction of a production burden and the proposition that the plaintiff still bears the burden of proof. Indeed, one must wonder why a federal court would puzzle over this issue because that is exactly what Fed.R.Evid. 301 provides.

Rule 301 was adopted by the Congress after it rejected a version proposed by the Advisory Committee and submitted to Congress by the Supreme Court. The Committee's version would have shifted the burden of persuasion to the adversary upon a party's establishing the basic facts that trigger a presumption. As adopted,1 Rule 301 shifts only the burden of production of evidence to rebut or meet the presumption, leaving the burden of persuasion on the party asserting the presumption. However, in adopting that Rule, the Conference Report expressly noted that, even after the adverse party has carried the burden of production, "the jury ... may [still] infer the existence of the presumed fact [unlawful discrimination] from proof of the basic facts [the McDonnell Douglas prima facie case]." H.R. Conf. Rep. No. 93-1597, at 2 (1974), reprinted in 1974 U.S.C.C.A.N. 7098, 7099; see also Mueller & Kirkpatrick, supra, § 3.8; 1 Stephen A. Saltzburg & Michael M. Martin, Federal Rules of Evidence Manual 99-100 (5th ed.1990); Weinstein et al., supra, p 301; 21 Charles Alan Wright & Kenneth W. Graham, Jr., Federal Practice and Procedure § 5126, at 606 (1977) & 1996 Supp. at 324.2

In short, under Rule 301, once the adversary satisfies the burden of production and the presumption "drops out," the prima facie case drops from version (i) to version (ii). Therefore, unless the Title VII/ADEA defendant's evidence of a lawful explanation is so powerful that the plaintiff's evidence is no longer sufficient to create a genuine dispute of fact, see Burdine, 450 U.S. at 255, 101 S.Ct. at 1094-95, the McDonnell Douglas prima facie case is sufficient to support (not compel) a finding of discrimination.3

The second principal reason given by my colleagues in the majority for the extinction of the plaintiff's prima facie case once a Title VII/ADEA defendant proffers an explanation for an adverse employment decision is that the term prima facie case has a "quite different" meaning in Title VII/ADEA cases. In their view, version (ii) of a prima facie case is not subsumed within version (i) when the McDonnell Douglas prima facie case is proven by the plaintiff. Version (ii) instead allows the plaintiff to complete the plaintiff's main case without legally sufficient evidence of either discrimination or causation, necessary elements they regard as absent from the McDonnell Douglas factors.

This admittedly "quite different" meaning of prima facie case is said to be designed to prevent the employer from remaining mute about its motives for an adverse employment decision and moving for summary judgment or a directed verdict when the plaintiff is unable to provide evidence of an unlawful motive. This purpose may be the source of the Title VII/ADEA prima facie case/pretext construct that has created so much confusion.

If the Supreme Court were to undertake a reexamination of this area, it might well consider whether its prior decisions with respect to the alteration of burdens were necessary, much less worth the enormous confusion they have caused. The Federal Rules of Civil Procedure entitle a plaintiff to take discovery of the defendant. The plaintiff is entitled to demand that the defendant furnish the reason for the adverse employment decision that is the subject of the suit. A defendant who refuses to answer such a discovery demand would be subject to sanctions forbidding it from contesting allegations of discriminatory motive and causation. See Fed.R.Civ.P. 37(b)(2). In reality, therefore, there is no such thing as a silent defendant, or if there is, it is a losing defendant. I therefore question whether the framework created by McDonnell Douglas and Burdine has furnished anything of value to weigh against the enormous confusion it has caused. The Supreme Court might do well to consider abandoning the entire experiment and rule explicitly that the burdens in a discrimination case are no different from those in other cases.4

In any event, even if the risk of a silent defendant calls for a burden of production, it does not follow that prima facie has a "quite different" meaning in Title VII/ADEA. As noted, Fed.R.Evid. 301 provides that all presumptions shift the burden of production but not the burden of proof and, as the Conference Report explicitly states, once the burden of production is met, the jury may be instructed that it "may infer the existence of the presumed fact from proof of the basic facts."

I then come to the crux--to the question of whether evidence satisfying the four factor test as illustrated in McDonnell Douglas is legally sufficient to allow a trier to find discrimination. The opinion of the court states with admirable candor that it does not. Judge Calabresi's concurring opinion agrees. Both opinions are based on the view of those judges that more is needed to support an inference of discrimination and causation than is required by McDonnell Douglas. I think that this conclusion is foreclosed by Supreme Court decisions.

The fact that the Court has consistently used the term prima facie case in Title VII cases surely cuts strongly against the majority's conclusion. The opinion for the court concedes that the term usually refers to evidence legally sufficient to support a particular finding. If the Supreme Court is in fact using the term in a "quite different" way, some express recognition of the linguistic departure would be expected, but none has come forth. The majority's view is impliedly that the Court seems to have misspoken, and on repetitive occasions.

Courts do misspeak, and the chance that it has happened here can't be dismissed. However, the Supreme Court has done more than simply use the term "prima facie." As noted, in Burdine, the Court not only used the term prima facie case but also equated it with evidence of "circumstances giving rise to an inference of discrimination," adding a footnote flatly stating that the four factor test as illustrated in McDonnell Douglas was sufficient to support just such an inference. Burdine, 450 U.S. at 253, n. 6, 101 S. Ct. at 1094 n. 6. Indeed, the footnote held that because the plaintiff in Burdine had met the McDonnell Douglas test, she had produced enough evidence to support an inference of discrimination. Similarly, in Int'l Bhd. of Teamsters, 431 U.S. 324, 97 S. Ct. 1843, 52 L. Ed. 2d 396 (1977) the Court specifically described a McDonnell Douglas prima facie case as "evidence adequate to create an inference that an employment decision was based on a discriminatory criterion." 431 U.S. at 358, 97 S.Ct. at 1866. This statement, which unqualifiedly asserts that a McDonnell Douglas prima facie case is sufficient to support an inference of discrimination and causation, was quoted with emphasis in O'Connor v. Consolidated Coin Caterers Corp., --- U.S. ----, ----, 116 S. Ct. 1307, 1310, 134 L. Ed. 2d 433 (1996). I do not believe that a Court of Appeals should, or even can, hold that the four factor test illustrated in McDonnell Douglas is not by itself sufficient to support an inference of discrimination and causation in light of the statements to the contrary in O'Connor, Burdine, and Teamsters.

2. A Prima Facie Case and the Proffer of a Pretext

My colleagues in the majority also hold the following: (1) a Title VII/ADEA plaintiff's presentation of a McDonnell Douglas prima facie case will, if the defendant stands mute as to the reasons for the adverse employment decision, result in liability; but, (2) the very same plaintiff's case can, if the defendant responds with a lie, be dismissed on insufficiency grounds. I take it that the majority would now disapprove the standard charge found in model jury instructions, 3 Edward J. Devitt et al., Federal Jury Practice and Instructions § 104.04 (1987); 5 Leonard B. Sand et al., Modern Federal Jury Instructions p 87.01 (Instruction 87-27) (1996), that the jury may (not must) infer a discriminatory motive from the consciousness of guilt reflected by dishonest statements about an adverse employment decision.

This view leads to peculiar results. For example, under the majority's reasoning, a prima facie case plus the proffer of a pretextual defense is weaker (from the plaintiff's standpoint) than an unanswered prima facie case. Moreover, it leaves Title VII/ADEA plaintiffs worse off than if McDonnell Douglas had provided for a version (ii) prima facie case than, as it did, for a version (i).

This puzzling ruling is also at odds with a vast body of law allowing inferences of consciousness of guilt to be drawn from dishonest behavior concerning facts material to litigation. For example, it is--or was until now--settled law that a false exculpatory statement by a defendant may (not must) support an inference of consciousness of guilt. See United States v. Sureff, 15 F.3d 225, 227 (2d Cir.1994); United States v. Gaviria, 740 F.2d 174, 184 (2d Cir.1984); United States v. Parness, 503 F.2d 430, 438 (2d Cir.1974); United States v. Lacey, 459 F.2d 86, 89 (2d Cir.1972); 1 Devitt et al., supra, § 14.06; 1 Sand et al., supra, p 6.05 (Instruction 6-11). Binder v. Long Island Lighting Co., 57 F.3d 193, 200 (2d Cir.1995), did no more than apply this universally recognized principle to false exculpatory statements by employers. My colleagues hold that this principle does not apply to employers or, perhaps, that it no longer applies to any party to litigation, including criminal defendants.

For another example, one of the most routine of jury instructions--perhaps given hundreds of times by some of my colleagues in the majority who are former district judges--states that if a jury finds that a witness has lied in a material part of his or her testimony, the jury may (not must) disbelieve other material parts of that witness's testimony. 1 Devitt et al., supra, § 15.06; 1 Sand et al., supra, p 7.01 (Instruction 7-19). Do such false statements now "point[ ] nowhere?" Fisher v. Vassar College, 70 F.3d 1420, 1437 (2d Cir.1995).

Similar instructions have been routinely given with regard to flight from the scene of a crime, 1 Devitt, et al., supra, § 14.08; 1 Sand et al., supra, p 6.05 (Instruction 6-9), use of a false name, 1 Sand et al., supra, p 6.05 (Instruction 6-10), fabrication of an alibi, id. (Instruction 6-12), use of disguised handwriting, id. (Instruction 6-13), falsification of evidence, 1 Devitt et al., supra, § 14.07; 1 Sand et al., supra, p 6.05 (Instruction 6-14), intimidation of witnesses, 1 Devitt et al., supra, § 14.07; 1 Sand et al., supra, p 6.05 (Instruction 6-16), and engaging in clandestine behavior, 1 Sand et al., supra, p 6.06 (Instruction 6-19).

Finally, in a criminal case, a defendant who takes the stand to deny the crime waives a sufficiency challenge to the prosecution's main case considered alone. United States v. Khan, 53 F.3d 507, 515 (2d Cir.1995); United States v. Friedman, 998 F.2d 53, 57 (2d Cir.1993); United States v. Roldan-Zapata, 916 F.2d 795, 803 (2d Cir.1990), cert. denied, 499 U.S. 940, 111 S. Ct. 1397, 113 L. Ed. 2d 453 (1991). Nevertheless, my colleagues in the majority hold that under Title VII/ADEA, an employer who proffers a lie about an adverse employment decision, can, precisely as a result of the act of lying, renew a previously rejected sufficiency challenge to a plaintiff's main case. Given that the prosecution's burden of proof in a criminal case is heavier than that of the Title VII/ADEA plaintiff, this vastly different treatment of pretextual defenses seems inexplicable unless my colleagues are prepared to change the rule now prevailing in criminal cases.

The majority reasons at length that a trier of fact's finding of discrimination is "reviewable." I agree. A decision that a prima facie case has been made out in either sense (i) or (ii) is of course reviewable by an appellate court, or trial court for that matter. With regard to version (i), the reviewable question is whether the evidence presented by the plaintiff is legally sufficient to allow the trier to find the basic facts that trigger the rebuttable presumption. With regard to (ii), the weaker version, the reviewable question is whether the plaintiff's evidence is legally sufficient to permit a trier to find discrimination. If the evidence is not legally sufficient, then no prima facie case is established and a finding of discrimination is "reviewable," i.e., clearly erroneous. Moreover, the employer may proffer evidence of a legitimate reason for the adverse employment decision so powerful that the plaintiff's prima facie case is undermined as a matter of law and is no longer sufficient to create a material dispute of fact. Burdine, 450 U.S. at 255, 101 S.Ct. at 1094-95.

A pretext finding may also be reviewed and found to be clearly erroneous. St. Mary's, 509 U.S. at 524, 113 S. Ct. at 2755-56. Finally, as Binder explicitly stated, even if the proffered reason is pretextual, an employer may present evidence explaining the resort to the pretext and giving a legitimate reason for the adverse employment decision. 57 F.3d at 200. Again, that evidence may be so powerful that the inference to be drawn from the prima facie case and resort to pretext are undermined as a matter of law.

What cannot be logically said is that a prima facie case in sense (ii) has been established and sustained but that a finding of discrimination is clearly erroneous. One might as well say that the evidence was legally sufficient to allow the trier of fact to find discrimination, but that the trier of fact's finding of discrimination is clearly erroneous because the evidence was not legally sufficient. Much less can one logically say that a finding of discrimination based on a prima facie case and supported by an unexplained resort to pretext is clearly erroneous.

I therefore dissent.