Nelson A. McCall appeals from a sentence imposed by Judge Scheindlin after he pleaded guilty to embezzlement. Judge Scheindlin calculated McCall's Sentencing Guidelines offense level based in part on her determination that a two-level "vulnerable victim" enhancement applied. She then sentenced McCall to a fifteen month term of imprisonment. McCall contends on appeal that Judge Scheindlin erred in applying the vulnerable victim enhancement. We remand for further findings.

McCall worked as a customer service representative at a branch of Apple Bank from September 1995 until February 27, 1997. In February 1997, an accountholder contacted the bank after receiving a form from the Internal Revenue Service indicating that an unauthorized distribution had been made from an Individual Retirement Account ("IRA") in the customer's name. Upon investigation, the bank discovered that the funds in question had been transferred to a savings account held by McCall. An investigation revealed that McCall had embezzled at least $196,976 from various accounts at the bank.

McCall pleaded guilty to embezzlement by a bank employee, in violation of 18 U.S.C. § 656. His presentence report included a two-level enhancement of his Guidelines offense level on the ground that McCall knew or should have known that one or more victims of his offense were particularly susceptible to the criminal conduct. See U.S.S.G. § 3A1.1(b) (vulnerable victim enhancement).

At McCall's sentencing hearing, the only contested issue was whether the facts warranted application of Section 3A1.1(b). McCall had embezzled primarily from IRAs, and many of the accountholders he targeted were over seventy years old. One was over ninety. Several were dead. In addition, among the accounts McCall had selected were pass-book accounts, inactive accounts, no-mail accounts,1 and accounts with bad addresses. Selecting accounts with these characteristics may have seemed to McCall to increase the chances that his activity would go undetected, although embezzlement through unauthorized withdrawals followed by deposits in the embezzler's savings account is anything but the perfect crime.

The district court found that McCall "knew or should have known that the victims he chose were particularly susceptible to the criminal conduct because of a combination of circumstances, including [but not limited to] their age." Other than age, it was the account characteristics described above that, the court found, made those McCall targeted "particularly susceptible" to his conduct. It noted:

These accounts from which he did embezzle ... a number of them were passbook accounts, which means that there was not regular mail, but you had to come in to have entries made in your passbook. And it seems to me logical that those people are less likely than those receiving statements to notice activity.

The court drew similar inferences with respect to the other types of accounts McCall targeted. As to the inactive, no-mail, and bad-address accounts, it found that "[McCall] did know [of the significance of these factors], and that is indeed why he singled these accounts out, because he knew that these folks were not going to get mail.... He could infer that this was a safer bet to target...."

The district court placed considerable weight on McCall's position at the bank, noting that "[h]e was an account representative. It strikes me that he would know which were the right accounts to target." In addition, the Presentence Report, the facts of which the court adopted, indicated that bank representatives had ready access to considerable information on IRA accountholders, and Section 3A1.1(b) expressly permits the attribution of constructive knowledge to defendants. Together with the age of those McCall targeted, the foregoing led the district court to conclude that his victims were "particularly susceptible" to his criminal activity and that he knew or should have known this. It then sentenced him, inter alia, to fifteen months imprisonment--the minimum sentence permissible under the applicable range.

We review a district court's findings of fact for clear error, and accord deference to its application of the Guidelines to the facts. See 18 U.S.C. § 3742(e); United States v. Borst, 62 F.3d 43, 46 (2d Cir.1995). The deference due " 'will depend upon the relationship of the facts to the guidelines standard being applied.' " United States v. Stroud, 893 F.2d 504, 506 (2d Cir.1990) (quoting 134 Cong. Rec. H11257 (daily ed. Oct. 21, 1988)). Where the particular determination closely resembles a finding of fact, we apply the clearly erroneous standard. As the inquiry approaches a purely legal determination, however, our scrutiny increases. See United States v. Castagnet, 936 F.2d 57, 59 (2d Cir.1991).

Section 3A1.1(b) provides, in pertinent part, that a defendant's offense level should be increased by two points if he "knew or should have known that a victim of the offense was unusually vulnerable due to age, physical or mental condition, or that a victim was otherwise particularly susceptible to the criminal conduct." U.S.S.G. § 3A1.1(b) (1997). In determining whether this provision applies, "[t]he courts appear to have interpreted 'susceptible to the criminal conduct' as emphasizing that a particular victim was less likely to thwart the crime, rather than more likely to suffer harm if the crime is successful." United States v. Kaye, 23 F.3d 50, 54 (2d Cir.1994); see United States v. O'Neil, 118 F.3d 65, 75 (2d Cir.) ("[F]ocus not on the likelihood or extent of harm to the individual if the crime is successful, but on the extent of the individual's ability to protect himself from the crime."), cert. denied sub nom. Saia v. United States, --- U.S. ----, 118 S. Ct. 728, 139 L. Ed. 2d 666 (1998); Borst, 62 F.3d at 46 (quoting Kaye ).

Some limits on the kinds of conduct that will justify a vulnerable victim enhancement are implied in Application Note 2 to Section 3A1.1(b). It states, inter alia, that the enhancement would not apply to someone who sold fraudulent securities by mail to the general public merely because one of his victims happened to be senile. See U.S.S.G. § 3A1.1(b) (Application Note 2). The securities fraud example is suggestive of two limits courts have generally read into Section 3A1.1(b). The first is that the vulnerability of the victim must bear some nexus to the criminal conduct. See, e.g., United States v. Monostra, 125 F.3d 183, 188-91 (3d Cir.1997) (remanding for nexus inquiry where corporate executive defrauded company whose president was blind). Thus, "[t]he adjustment would apply ... in a robbery where the defendant selected a handicapped victim," U.S.S.G. § 3A1.1(b) (Application Note 2), but might not apply in a fraud case involving the same physically handicapped person. The second limit suggested by the securities fraud example is that the defendant generally must have singled out the vulnerable victims from a larger class of potential victims. Compare United States v. Stover, 93 F.3d 1379, 1388 (8th Cir.1996) (reversing enhancement in adoption fraud case because defendant targeted anyone willing to pay his fees), and United States v. Wilson, 913 F.2d 136, 138 (4th Cir.1990) (reversing enhancement where defendant mailed fraudulent solicitation letters to randomly selected P.O. boxes), with United States v. Holmes, 60 F.3d 1134, 1137 (4th Cir.1995) (affirming enhancement in fraud case where defendant singled out people with poor credit histories); see also United States v. Singh, 54 F.3d 1182, 1191 (4th Cir.1995) ("[I]n a proper § 3A1.1 enhancement the victim must be more susceptible to abuse from a perpetrator than most other potential victims of the particular offense."); United States v. Caterino, 957 F.2d 681, 683 (9th Cir.1992) (reading securities fraud example as excluding cases where defendant does not know of victim's vulnerability).

However, the securities fraud example notwithstanding, Section 3A1.1(b) does not require that the defendant select the victim because of his or her vulnerability--it is sufficient that he knew or should have known of this quality when deciding to go ahead with the crime. See United States v. Gill, 99 F.3d 484, 488 (1st Cir.1996) (noting 1995 amendment to Section 3A1.1 eliminating "previously required element of targeting motivation").

By contrast, a third limit, imposed by courts, is that broad generalizations about victims based upon their membership in a class are disfavored where a very substantial portion of the class is not in fact particularly vulnerable to the crime in question. In such cases, courts have required that the enhancement be based on individualized findings as to the vulnerability of particular victims. See, e.g., United States v. Fosher, 124 F.3d 52, 56 (1st Cir.1997) (remanding for finding as to whether 62-year-old victim of home invasion was unusually vulnerable). It is the concern over the lack of individualized findings that has given rise to the proposition that age alone is insufficient to support the enhancement. See, e.g., O'Neil, 118 F.3d at 75; United States v. Lee, 973 F.2d 832, 835 (10th Cir.1992) (reversing vulnerability finding based solely on victims' age).

Nevertheless, although there is skepticism of generalized assumptions about a victim's vulnerability based upon that person's membership in a class, many cases have upheld vulnerable victim enhancements based on group generalizations. See, e.g., United States v. Echevarria, 33 F.3d 175, 180-81 (2d Cir.1994) (medical patients); United States v. Patasnik, 89 F.3d 63, 72 (2d Cir.1996) (people with poor credit histories). Indeed, Application Note 2 makes clear that class membership alone may be sufficient to support the enhancement by stating that someone who "marketed an ineffective cancer cure" would warrant the enhancement. In such cases, however, the criminal shapes the nature of the crime--fraudulent cancer cure rather than fraudulent land sales--to target a class of victims that are virtually all particularly vulnerable to that crime.

In the present case, the district court concluded that McCall targeted accountholders that he knew to be less likely than most to detect his criminal conduct. In light of McCall's position at the bank and because constructive knowledge suffices under Section 3A1.1(b), this conclusion is entirely reasonable. McCall's contention, however, is that generalized information about accounts cannot be a basis for a vulnerability finding with respect to particular accountholders. We disagree with so sweeping a view of vulnerability but remand to the district court because of two concerns with its application of the vulnerable victim enhancement.

Our first concern is that the district court may have applied an incorrect legal test. It seemed to state at sentencing that only someone who is more susceptible to a crime than most victims of that crime meets the particularly vulnerable standard. In particular, it noted that holders of passbook accounts are "less likely than those receiving statements to notice activity" and that no-mail and bad-address accounts were "a safer bet to target." However, a more vulnerable-than-most test varies considerably from a particularly vulnerable test. See Stover, 93 F.3d at 1386 ("courts have consistently refused to find a class of victims to be particularly susceptible ... simply because they were statistically more likely to fall prey to the defendant's crime.' " (quoting United States v. Castellanos, 81 F.3d 108, 110-11 (9th Cir.1996))). To hold otherwise would be to read the term "particularly" out of Section 3A1.1(b).

Moreover, the more-vulnerable-than-most test would lead to an enhancement whenever any member of the most vulnerable 49% + of the potential class of victims is selected by the criminal but not when a member of the 50% + less vulnerable is targeted. This result is in some cases overinclusive, in others underinclusive. Some crimes--armored car robberies--are unlikely to involve a particularly vulnerable victim, although some armored car companies may be more vulnerable than others. Other crimes--fraudulent cancer cures--may be directed at a class of victims who are virtually all particularly vulnerable. The correct test calls for an examination of the individual victims' ability to avoid the crime rather than their vulnerability relative to other potential victims of the same crime. The enhancement, in short, is to be applied where an extra measure of deterrence and punishment is necessary because the defendant knew of a victim's substantial inability to avoid the crime. See Stover, 93 F.3d at 1386-87 (" [T]he victims to whom § 3A1.1 applies are those who are in need of greater societal protection.... They are the persons who, when targeted by a defendant, render [his] conduct more criminally depraved." (quoting Castellanos, 81 F.3d at 110-11)).

Our second concern is with the facts relied upon to apply the vulnerable victim enhancement to appellant. In the circumstances of this case, the bank rather than the accountholder is liable for an embezzlement. However, accountholders are nevertheless victims of such an embezzlement. They are put at risk and will at the least suffer substantial inconvenience. Such an accountholder may be a particularly vulnerable victim where there is a substantial chance that he or she will never discover or realize that the account has been depleted. Such an accountholder may be particularly vulnerable where the nature of the embezzlement renders it difficult for the accountholder to establish that the withdrawals were unauthorized, causing at the best inconvenience and anxiety and at the worst a temporary need for, or loss of, funds.

In the present case, the district court relied upon, without further particularized inquiry, the ages of some accountholders and other characteristics of the accounts McCall targeted. In using this generalized analysis, the district court appears to have concluded that each of the accountholders in question was a particularly vulnerable victim because every account featured at least one of these characteristics, e.g., elderly accountholders, pass-book accountholders, etc. However, none of these characteristics alone demonstrates a particular vulnerability to embezzlement. For example, elderly people can be meticulous about their finances. Cf. Lee, 973 F.2d at 835 (simple fact that victims were elderly not sufficient to support Section 3A1.1). Even the fact that someone has a pass-book account is not a sufficient basis by itself for the enhancement, because the accountholder may in fact keep track of the account. Likewise, the fact that an accountholder is deceased, without more, does not render his or her estate particularly vulnerable to this sort of embezzlement.

It may well be the case that one or more of McCall's intended victims was indeed particularly susceptible to his conduct because of age and other factors. Cf. United States v. Smith, 133 F.3d 737, 749 (10th Cir.1997), cert. denied, --- U.S. ----, 118 S. Ct. 2306, 141 L. Ed. 2d 165 (1998) ("A single vulnerable victim is sufficient."). However, the district court did not detail the combination of characteristics of individual accounts--for example, a passbook account in the name of a dead person in which there has been no activity for several years--that rendered individual accountholders particularly vulnerable.

We therefore remand to the district court to clarify the record as to which accountholders were particularly vulnerable and what combination of factors made them so. The district court has discretion to allow supplementation of the record as it deems appropriate. It may also, of course, entertain any application for bail during pendency of this appeal. The mandate shall issue forthwith and any party seeking appellate review of the decision on remand shall so inform the clerk of this court within 30 days of that decision. Jurisdiction will then be automatically restored to this court. See United States v. Jacobson, 15 F.3d 19, 21-22 (2d Cir.1994). After jurisdiction is restored, the clerk shall set an expedited schedule for letter briefs, and the matter will then be heard by this panel.

We remand for further proceedings consistent with this opinion.