FOR PUBLICATION
 UNITED STATES COURT OF APPEALS
      FOR THE NINTH CIRCUIT

EQUAL EMPLOYMENT OPPORTUNITY         
COMMISSION,
                                           No. 07-16903
              Plaintiff-Appellant,
              v.                            D.C. No.
                                         CV-05-03034-FJM
THE BOEING COMPANY, a Delaware
                                            OPINION
corporation,
             Defendant-Appellee.
                                     
       Appeal from the United States District Court
                for the District of Arizona
      Frederick J. Martone, District Judge, Presiding

                 Argued and Submitted
             March 6, 2009—Tucson, Arizona

                  Filed August 18, 2009

  Before: Michael Daly Hawkins, Marsha S. Berzon, and
           Richard R. Clifton, Circuit Judges.

                Opinion by Judge Hawkins




                          11237
11240         EEOC v. THE BOEING COMPANY


                      COUNSEL

Anne Noel Occhialino (authored briefs and argued), Equal
Employment Opportunity Commission, Washington, D.C.,
for the plaintiff-appellant.
                EEOC v. THE BOEING COMPANY              11241
Tibor Nagy, Jr. (argued) and Erica K. Rocush (authored
brief), Ogletree, Deakins, Nash, Smoak & Stewart, Tucson,
Arizona, for the defendant-appellee.


                         OPINION

HAWKINS, Circuit Judge:

   The Equal Employment Opportunity Commission
(“EEOC”) appeals, on behalf of charging parties Antonia
Castron (“Castron”) and Renee Wrede (“Wrede”), the grant of
summary judgment to Boeing in this action under Title VII of
the Civil Rights Act of 1964. Boeing terminated Castron and
Wrede after they received low scores on reduction-in-force
(“RIF”) assessments, which Boeing uses to evaluate employ-
ees when determining whom to lay off. We hold that the
EEOC introduced adequate evidence from which a reasonable
jury could conclude that the reasons Boeing advanced to jus-
tify its employment actions were pretextual. Accordingly, we
reverse and remand for a trial on both charging parties’ dis-
crimination claims and Castron’s retaliation claim.

                     BACKGROUND

  The following recitation of the facts reflects the nonmoving
party’s factual submissions, which we credit for purposes of
summary judgment. See Vasquez v. County of Los Angeles,

, 639-40 (9th Cir. 2003).

Antonia Castron

  Castron, after working for Boeing sporadically for several
years, joined its Electrical Engineering Department as a liai-
son engineer in 1997. From 1998 until her termination in
2003, Castron worked under department manager Bill Charl-
ton (“Charlton”), who reviewed the performance evaluations
given by supervisors to liaison engineers.
11242           EEOC v. THE BOEING COMPANY
   According to the deposition testimony of Boeing employee
Glen Foster (“Foster”), Charlton had frequently made nega-
tive comments about women, including comments to the
effect that Charlton “didn’t want any more women and that
women were not worth a shit,” that “he didn’t have good luck
with females and they hadn’t been around long enough for his
satisfaction,” that he “just didn’t have time” to train women,
and that Charlton’s ex-wife, who was a Boeing employee,
“should be at home, not working.” According to Foster, Char-
lton, in the context of discussing a female candidate for a pro-
duction engineer position, said he “basically felt [women]
should come to him all trained,” an expectation Foster had
never heard him express with regard to men.

   According to her sworn affidavit, in 2001, Castron began
to feel mistreated and unwelcome by her exclusively-male
coworkers in Boeing’s Pre-Shop Analysis (“PSA”) depart-
ment because she was a woman. Castron reported her percep-
tion to several supervisors and repeatedly requested a transfer.
In November 2001, Castron asked Charlton to transfer her to
the Final Assembly workgroup. At the suggestion of another
supervisor, Sam Turk (“Turk”), Castron told Charlton that she
wished to transfer for career advancement, rather than
because of harassment. Turk also asked Charlton several
times to transfer Castron to Final Assembly.

   Despite these repeated requests, Charlton refused to trans-
fer Castron to the Final Assembly workgroup, but instead
transferred a male coworker. When Castron sought an expla-
nation, Charlton responded angrily that he would transfer her
in three months. On August 6, 2002, after Castron told Charl-
ton that the “hostile work environment” in the PSA depart-
ment was “keeping [her] from performing [her] job,” Charlton
proposed transferring Castron not to the Final Assembly
department she requested, but instead to the Structures-Mod
department. No transfer of that type had occurred in the previ-
ous four years because the Structures-Mod department
                 EEOC v. THE BOEING COMPANY                11243
required substantially different skills from those required in
the PSA department.

   As Charlton admitted in his EEOC statement, Castron was
“reluctant” to be transferred from PSA to Structures-Mod and
thought Joe Cottone, a male-employee with whom Castron
had been in conflict, “should move and not her.” Charlton’s
EEOC statement noted, however, that “[a]fter further discus-
sion, we all agreed that [Castron] should move.”

   Although Castron did ultimately agree to leave PSA, she
had two major concerns. First, she was concerned about
working for Rick Hobby (“Hobby”), who would be her super-
visor in the Structures-Mod department, because of previous
incidents in which he had referred to Castron as a “little girl,”
joked about Castron breaking a nail, and perfunctorily apolo-
gized and stormed off when confronted. Second, Castron
feared the transfer to Structures-Mod would make her vulner-
able to firing in an upcoming RIF assessment. Castron stated
that Structures-Mod work is more difficult and complex than
the work in PSA and was outside of Castron’s core area of
expertise. Castron agreed to the transfer only after Charlton
assured her that the upcoming RIF would not affect her in
Structures-Mod. Castron’s coworkers suggested such exemp-
tions did occur and that this promise was plausible.

   Just two months after Castron’s transfer, Boeing conducted
a RIF, in which employees were evaluated in several catego-
ries and those with the lowest scores were eligible for termi-
nation. Hobby prepared the final scores for all employees
subject to the RIF. Despite Charlton’s past assurances,
Castron was subject to the RIF, received low scores, and was
ultimately terminated as a result.

   Hobby’s evaluation of Castron included low scores for
“ability to perform remaining work” and “past performance.”
Hobby contended that the scores were based on personal
interactions and comments from senior engineers, although
11244           EEOC v. THE BOEING COMPANY
Hobby had little personal interaction or knowledge of
Castron’s performance. According to Turk, “past perfor-
mance” scores for a recently-transferred employee typically
include an employee’s entire body of work. Nevertheless,
Hobby stated that he based Castron’s “past performance”
scores solely on her two months as a trainee in Structures-
Mod.

   Leendert Hartoog (“Hartoog”), who trained and supervised
Castron in Structures-Mod, stated that Castron had made good
progress in her new position, but noted that typical training
time for someone with Castron’s background was five or six
years. Hartoog stated that he expected Castron to receive
higher scores or to receive exemptions from certain require-
ments unfair to impose on a trainee, and that he believed
Castron was “set up to fail.”

Renee Wrede

  Wrede began working for Boeing in 1989. In 1999, after
Boeing substantiated Wrede’s complaint of sexual harassment
by her direct supervisor, Wrede was transferred to the Apache
helicopter manufacturing assembly installation support group.
There, Wrede reported to Bruce Wright (“Wright”), who
reported in turn to Rob Feuerstein (“Feuerstein”), the depart-
ment manager for the division.

   In 2001, Wright gave Wrede a positive evaluation, rating
her as “meets expectations” or higher in ten categories and
between “below expectations” and “meets expectations” in
two other categories. The evaluation noted only one “positive
flaw,” which was that Wrede saw herself as “Miss Fix-It” and
should have “passed off responsibilities when appropriate
more often rather than trying to do it all” herself. Wright
placed Wrede on a “performance improvement plan,” which
Wright testified was intended to improve the two areas in
which she received the lowest evaluation scores.
                EEOC v. THE BOEING COMPANY                11245
   In April 2002, Boeing conducted a RIF for manufacturing
engineers (“MEs”), which required Wright and Feuerstein to
evaluate Wrede again. Wrede scored thirtieth out of thirty-
seven MEs. Although Boeing never completed the RIF
because several MEs transferred or left the company, Wrede’s
evaluation did not place her in the band eligible for termina-
tion.

   In July 2002, Boeing conducted another RIF for employees
in Wrede’s skill code, but based the RIF primarily on previ-
ous RIF scores. Although the four lowest-ranking MEs, all of
whom were men with scores lower than Wrede’s scores, were
initially slated for termination, again, none was actually laid
off. One had his RIF cancelled and remained, two transferred
to another division, and the fourth had his RIF cancelled
because he took another ME job with Boeing in Texas.

   In October 2002, Boeing conducted a third RIF for Wrede’s
skill code. Wright performed Wrede’s RIF assessment. He
scored Wrede significantly lower in her “ability to perform
remaining work” in this evaluation than in the earlier evalua-
tions, particularly in the “communication/leadership”category,
which measured “team skills,” “communication skills,” and
“leadership skills.”

   Wright testified the lower scores were justified by a decline
in team-building, communication, and leadership skills,
although Wright was unable to point to any specific com-
plaints or any written record of problems. Wright and Feuer-
stein noted that Wrede missed deadlines and overloaded
herself with work, although Wrede testified that other MEs
also failed to meet deadlines and that Wright never raised the
issue with Wrede. In addition to Wrede’s ability to perform
remaining work, Wrede received the second lowest score for
“past performance,” but Wright was unable to explain why
Wrede’s scores were so low.

  Wrede also received lower technical scores in her third RIF
evaluation. Wrede received a “0,” reflecting “no knowledge”
11246           EEOC v. THE BOEING COMPANY
or “no experience” in multiple categories, such as “Unigraph-
ics” and “wiring systems,” even though her own statements
and prior RIF evaluations had indicated she possessed at least
some knowledge or experience in these areas.

   After Feuerstein met with Wright and the other managers
to ensure uniform scoring, six men and Wrede, the only
female ME in her skill code, were given RIF notices. Unlike
Wrede, however, all six low-scoring male engineers ulti-
mately avoided termination. One of them, James Early, had
his notice cancelled after expressing concerns regarding his
assessment process. Five others avoided layoffs by transfer-
ring within Boeing, in two cases, because Feuerstein told
them about new job openings, assistance Wrede stated Feuer-
stein never offered her.

  In a sworn affidavit, David Eroh (“Eroh”), a lead ME in
Wrede’s skill code who obtained high RIF scores, stated that
Wrede was better than “many of the other MEs,” equaled his
own capabilities, and should have received higher scores in
many categories. Eroh noted that Wrede “excelled where
other colleagues” failed, handled numerous jobs without
assistance, became a trusted assistant to management, and
produced “stellar” work. Eroh gave detailed explanations
why, in his opinion, each of Wrede’s low scores should have
been higher.

   Likewise, several managers with whom Wrede was work-
ing held her in high regard. Multiple managers stated that cer-
tain male employees should have been laid off instead or
asked Feuerstein if Wrede could be transferred to their work-
ing groups. Notwithstanding these statements and requests,
Feuerstein approved Wrede’s termination.

   The district court granted summary judgment to Boeing on
all claims, finding that the EEOC had failed to establish a
prima facie case for Castron’s discriminatory transfer claim
by identifying a similarly-situated male who was treated more
                EEOC v. THE BOEING COMPANY                11247
favorably and that the EEOC had failed to present sufficient
evidence to establish pretext with respect to any of the claims.
The EEOC timely appealed with respect to all discrimination
claims and Castron’s retaliation claim, but does not pursue
Wrede’s retaliation claim on appeal.

                STANDARD OF REVIEW

   We have jurisdiction pursuant to 28 U.S.C. § 1291 and
review de novo the district court’s grant of summary judg-
ment. See 

. Viewing the evidence in
the light most favorable to the nonmoving party, we must
determine whether there are any genuine issues of material
fact and whether the district court correctly applied the rele-
vant substantive law. 

                        DISCUSSION

   A plaintiff alleging disparate treatment under Title VII
must first establish a prima facie case of discrimination by
offering evidence that “give[s] rise to an inference of unlaw-
ful discrimination.” Tex. Dep’t of Cmty. Affairs v. Burdine,

, 253 (1981). A plaintiff may establish a prima
facie case either by meeting the four-part test laid out in
McDonnell Douglas Corp. v. Green, 

, 802
(1973), or by providing direct evidence suggesting that the
employment decision was based on an impermissible crite-
rion, Cordova v. State Farm Ins. Cos., 

, 1148
(9th Cir. 1997). Once a prima facie case has been made, “[t]he
burden of production, but not persuasion, then shifts to the
employer to articulate some legitimate, nondiscriminatory
reason for the challenged action.” Chuang v. Univ. of Cal.
Davis, 

, 1123-24 (9th Cir. 2000). If the
employer does so, the plaintiff must then show that the articu-
lated reason is pretextual “either directly by persuading the
[fact-finder] that a discriminatory reason more likely moti-
vated the employer or indirectly by showing that the employ-
er’s proffered explanation is unworthy of credence.” Burdine,
11248            EEOC v. THE BOEING 

. When the evidence is direct, “ ‘[w]e require
very little evidence to survive summary judgment’ in a dis-
crimination case.” Lam v. Univ. of Haw., 

, 1564
(9th Cir. 1994) (quoting Sischo-Nownejad v. Merced Cmty.
Coll. Dist., 

, 1111 (9th Cir. 1991)) (alteration
in original). “But when the plaintiff relies on circumstantial
evidence, that evidence must be specific and substantial to
defeat the employer’s motion for summary judgment.” Cogh-
lan v. Am. Seafoods Co. LLC, 

, 1095 (9th Cir.
2005) (internal quotation marks omitted).

I. Antonia Castron

   [1] After complaining of a hostile work environment,
Castron was transferred to a new work group and was termi-
nated in a RIF two months later. The critical inquiry is
whether Castron’s employment experience, including her
transfer and its connection to a subsequent RIF that led to her
termination, would allow a jury to find in favor of the EEOC.

   We conclude the EEOC has established a prima facie case
on Castron’s behalf because of direct evidence of discrimina-
tory animus. 

. In RIF cases, a
plaintiff can “show through circumstantial, statistical or direct
evidence that the discharge occurred under circumstances giv-
ing rise to an inference of . . . discrimination.” Nesbit v. Pep-
sico, Inc., 

, 705 (9th Cir. 1993) (per curiam).

   [2] Foster testified that Charlton, Castron’s supervisor, fre-
quently made demeaning and derogatory comments about
women. These comments, considered along with Charlton’s
interactions with Castron over the course of her employment
at Boeing, are sufficient to create an inference of discrimina-
tory motive even though the comments were not directed spe-
cifically at Castron or made in regard to decisions about her
employment. See 

; Dominguez-
Curry v. Nev. Transp. Dep’t, 

, 1038 (9th Cir.
2005); 

. These comments are more
                 EEOC v. THE BOEING COMPANY                11249
severe than “ambivalent” ”stray remark[s]” that we have pre-
viously held insufficient to establish such an inference. See,
e.g., 

; Nidds v. Schindler Elevator
Corp., 

, 918-19 (9th Cir. 1997). Consequently,
the EEOC made out a prima facie case in relation to Castron’s
transfer.

   [3] Because Boeing has articulated legitimate, nondiscrimi-
natory reasons for its decision to transfer and subsequently
terminate Castron—her request for a transfer and her subse-
quent low RIF scores—the EEOC was required to respond
with evidence from which a jury could infer that Boeing’s
proffered reasons are pretextual. 

.
The discriminatory animus exhibited by Castron’s supervisor
constitutes direct evidence of pretext, even though the com-
ments did not refer specifically to Castron. Based on Charl-
ton’s sexist comments, a jury might reasonably infer that
Charlton’s decision to transfer Castron, rather than a male
coworker about whom she complained, to a new position
where her job was less secure, may have resulted from
improper motivations, including discriminatory intent, retalia-
tory intent, or both. There is also “specific and substantial”
circumstantial evidence, in addition to Charlton’s comments,
on which a jury could rely to support the conclusion that the
transfer might have been discriminatory. A jury might credit
Castron’s allegations that Charlton (1) initially refused to
transfer Castron at all, (2) made promises to transfer her to the
department she requested, (3) agreed to transfer her, but only
to a different department to which no other engineers from her
department had been transferred in recent years, and (4)
assured Castron that she would be exempt from the RIF pro-
cess during her training in order to induce Castron to accept
the transfer despite her explicit (and not unwarranted) concern
that the transfer might significantly increase her risk of termi-
nation. Taking note of all of the direct and circumstantial evi-
dence, a jury might conclude that Charlton deliberately set
Castron up to fail because of her sex or because of her invoca-
tion of Title VII rights.
11250           EEOC v. THE BOEING COMPANY
   [4] There is also sufficient evidence from which a jury
could find that Castron’s later poor RIF evaluation scores,
which led to her termination, were pretextual. Castron’s
supervisor in her new department, Hobby, had previously
referred to Castron as a “little girl” and made a “joking”
inquiry as to whether she “broke a nail.” Although these com-
ments occurred two years prior to Castron’s firing and Boeing
argues these comments are mere “stray remark[s],” Hobby’s
comments constitute at least some evidence of discriminatory
animus.

   [5] Moreover, Hobby evaluated Castron without asking
Castron’s trainer about her progress. Several employees testi-
fied that Hobby unfairly ignored Castron’s past performance
evaluations and instead focused only on her two months as a
trainee in her new department, that Castron’s skills merited
higher scores, and that Hobby gave Castron lower scores than
those received by other male employees from Castron’s previ-
ous department who allegedly possessed skills inferior to
Castron’s.

   [6] Boeing urges us to consider this testimony by other
employees irrelevant because Castron’s coworkers’ views do
not prove that Hobby’s differing subjective evaluations were
either incorrect or pretextual. Although subjective evaluations
of an employee’s skills of course may differ for a variety of
reasons, specific positive evaluations of Castron’s perfor-
mance, both by her coworkers and by other managers, criti-
cally undermine the credibility of her official evaluation in a
manner relevant to determining the existence of pretext. We
therefore adopt the Tenth Circuit’s view that “co-workers’
assessment[s]” of a plaintiff’s work should be considered
because they can be “clearly probative of pretext.” Abuan v.
Level 3 Commc’ns, Inc., 

, 1174 (10th Cir.
2003).

   [7] In light of all of the evidence of pretext introduced by
the EEOC, a reasonable jury could infer that Hobby’s evalua-
                  EEOC v. THE BOEING COMPANY                     11251
tion of Castron was not worthy of credence and therefore pre-
textual. Because the EEOC has presented adequate direct
evidence and “specific and substantial” circumstantial evi-
dence that Castron’s termination was pretextual, entry of sum-
mary judgment was erroneous. A jury could find in favor of
the EEOC on all of Castron’s discrimination and retaliation
claims regarding both her transfer and termination.

II. Renee Wrede

   In October 2002, one year after Boeing substantiated a sex-
ual harassment claim Wrede had filed, she received lower RIF
scores than most engineers in her skill code and was subse-
quently terminated.1 These scores were lower than the scores
she had received in two previous RIF evaluations in April and
July of 2002. Wrede scored high enough on the earlier RIFs
to avoid vulnerability to discharge, but her scores in the Octo-
ber RIF dropped substantially, placing her at risk. Although
several male engineers were also initially selected for termi-
nation, none was ultimately terminated in any of the three
RIFs, because they either successfully contested their scores
or found other employment within Boeing, sometimes with
the assistance of their supervisors.

   [8] Both parties agree that the EEOC established a prima
facie case of gender discrimination regarding Wrede under
McDonnell Douglas, 

, that Boeing has articu-
lated a legitimate, non-discriminatory reason for terminating
Wrede (her low RIF scores), and that the EEOC has intro-
duced circumstantial, but not direct, evidence that this reason
for Wrede’s termination was pretextual. The parties dispute
only whether the EEOC presented sufficiently “specific and
substantial” circumstantial evidence that Wrede’s RIF scores
were not credible to allow a jury to find pretext.
  1
    The district court granted summary judgment on Wrede’s Title VII
retaliation claim because of the year that elapsed between her protected
conduct and the adverse employment action. The EEOC has not appealed
this decision. We therefore consider only Wrede’s discrimination claim.
11252            EEOC v. THE BOEING COMPANY
   [9] As it concedes, the EEOC must also overcome an infer-
ence arising from the fact that the same actors who made an
adverse employment decision against Wrede in the October
RIF had twice given her scores that were high enough to
avoid vulnerability to discharge. While recent positive
employment decisions made by the same actors who later
make an adverse employment decision against an employee
may give rise to “an inference that no discrimination took
place,” Coleman v. Quaker Oats Co., 

, 1288
(9th Cir. 2000), the inference in this case is weaker than the
“strong inference” against bias that arises when an employer
who hires or promotes a plaintiff later takes an adverse
employment action against her, see 

.

   [10] Although a termination is rarely motivated by bias
when it is initiated by the same actors who recently selected
the same employee for the job or promotion in the first place,
the logic differs when applied to less overtly “positive”
employment decisions, such as refraining from firing an
employee at the earliest opportunity or giving an employee a
lukewarm evaluation, rather than a poor one. A supervisor
who hires or promotes an employee affirmatively forwards
the employee’s career; this affirmative enhancement of the
employee’s career prospects is strong circumstantial evidence
of a lack of bias on the supervisor’s part. In contrast, where
a supervisor discharges an employee he did not affirmatively
hire or promote by lowering her scores over time, rather than
by firing her precipitously, there is no such strong circumstan-
tial evidence of lack of bias.

   [11] A strong inference is also inappropriate here because
none of the employees who the supervisors ranked lower than
Wrede in the April and July RIFs ultimately suffered as a
result: all six men had their RIFs cancelled or obtained trans-
fers within Boeing, and none was laid off. Given that a jury
could find the evidence suggested Wrede’s RIF scores were
not credible, as discussed further below, a jury could infer that
                EEOC v. THE BOEING COMPANY                11253
Wrede suffered discrimination notwithstanding any inference
arising from her supervisors’ prior employment decisions.

   [12] The EEOC has also produced evidence from which a
jury could conclude that Wrede’s RIF assessment was pretex-
tual. First, Wright assigned Wrede RIF scores in October indi-
cating “no background or experience” in areas in which she
had received higher scores in earlier RIFs, indicating at least
some background or experience. Wrede also received signifi-
cantly lower evaluations in several soft-skill categories, such
as “communication/leadership,” in the October RIF than in
the July RIF, even though Wright was unable to offer a non-
conclusory explanation of any of the significant changes or to
point to any concrete conduct, specific complaints, or written
records indicating a change. Wright also contended Wrede
had trouble communicating with her “dotted-line manager,”
but was unable to recall who Wrede’s dotted-line manager
was.

   Other specific and substantial circumstantial evidence also
suggests Wright lacked legitimate justification for his scoring.
Several of Wrede’s coworkers and managers offered detailed
testimony regarding why the RIF assessments of Wrede’s
skills were not credible. Coworkers’ and managers’ assess-
ments of an employee’s skills and performance can be proba-
tive of pretext. See 

. Coworker
testimony is particularly relevant here because it would allow
a jury to infer that Boeing’s proffered reason for termination
— a poor RIF evaluation — was not only inaccurate, but is
simply unworthy of credence.

   Specifically, coworkers and managers familiar with
Wrede’s work gave detailed testimony that Wrede was a good
employee, that her skills warranted higher scores than she
received, and that she performed better than several male sur-
vivors of the October RIF. Several managers even requested
that Feuerstein transfer Wrede to their departments, but
Feuerstein denied these requests. Furthermore, Wrede’s own
11254            EEOC v. THE BOEING COMPANY
detailed testimony about why her low scores were not merely
“wrong or mistaken,” but were “unworthy of credence,” is
also evidence for the jury to consider. Tomasso v. Boeing Co.,

, 708 (3d Cir. 2006) (internal quotation marks
omitted).

   [13] Finally, Wrede, the only woman in her skill code, was
laid off while every male employee identified for termination
in all three RIFs ultimately remained at Boeing, sometimes
due to the assistance of supervisors, assistance that was not
made similarly available to Wrede. Feuerstein’s failure to
treat Wrede the same as male employees and “the inexorable
zero” female employees who remained in the department after
the RIF are also probative of pretext. Int’l Bhd. of Teamsters
v. United States, 

, 342 n.23 (1977); see also

(“[S]tatistical evidence is circum-
stantial evidence that could, if sufficiently probative, point to
bias.”). All of this could lead a jury to conclude that Boeing’s
asserted rationales were pretextual.

   [14] Because the EEOC has presented specific and substan-
tial evidence in support of Wrede’s claim that Boeing’s
asserted justification for her termination was pretextual, sum-
mary judgment on Wrede’s discrimination claim was errone-
ous. A jury could reasonably conclude that Wrede’s discharge
resulted from discrimination on account of sex.

  For all of these reasons, the district court’s grant of sum-
mary judgment is reversed and the case is remanded for pro-
ceedings consistent with this Opinion.

  REVERSED AND REMANDED.