Janet Stout, Juliana Nedd, Sheila Wright and Lee Harrison (collectively, the “postal inspectors”) appeal the district court’s summary judgment in favor of the Postmaster General in their employment discrimination action alleging denial of promotion on the basis of sex. We have jurisdiction, 28 U.S.C. § 1291 , and we affirm.

I

Appellants are female postal inspector team leaders in the Postal Inspection Service (“the Service”), the law enforcement branch of the United States Postal Service. They, along with 34 other postal inspectors, applied for promotion to Assistant Inspector in Charge (“AIC”), the highest non-executive managerial level in the Service. There were five open AIC positions, one each in San Francisco, Los Angeles, Houston, Milwaukee and Washington, D.C. Six of the 38 applicants who vied for these positions were women.

A review panel initially screened all applicants on the strength of their supervisor evaluations and applications. The panel identified the most qualified candidates and forwarded their names as potential interviewees to a separate selection committee that made the final hiring decisions. From the original pool of 38, the screening panel identified 10 applicants as the most qualified. None of the six female applicants was named to this list and none was initially interviewed by the selection committee.

Two female applicants were granted interviews in a second screening round which arose from unexpected circumstances. The selected candidate from the first round of interviews for the San Francisco position declined an offer. The Inspector in Charge of that office was not satisfied with the remaining candidates who were first interviewed and asked the screening panel to select additional names from the original pool of applicants. Two of the additional five applicants chosen to be interviewed were female applicants. One of these female applicants ultimately was promoted to the San Francisco AIC opening.

The postal inspectors commenced this action alleging that the Service caused them to suffer both disparate treatment and disparate impact on the basis of sex in violation of Title VII of the Civil Rights Act of 1964, 42 U.S.C. §§ 2000e to e-17. The district court granted the Service’s motion for summary judgment on both claims of discrimination. The postal inspectors now appeal, contesting only the dismissal of their disparate impact claim.

II

We review a district court’s grant of summary judgment de novo. Lopez v. Smith, 203 F.3d 1122, 1131 (9th Cir.2000) (en banc). We determine, viewing the evidence in the light most favorable to the nonmoving party, whether there are any genuine issues of material fact and whether the district court correctly applied the relevant substantive law. Id.

III

A claim of disparate impact challenges “employment practices that are facially neutral in their treatment of different groups but that in fact fall more harshly on one group than another and cannot be justified by business necessity.” Int’l Brotherhood of Teamsters v. United States, 431 U.S. 324 , 335 n.15, 97 S.Ct. 1843 , 1854 n. 15, 52 L.Ed.2d 396 (1977); see also 42 U.S.C. § 2000e-2(k)(l)(A)(i). A plaintiff establishes a prima facie case of disparate impact by showing a significant disparate impact on a protected class caused by a specific, identified, employment practice or selection criterion. Wards Cove Packing Co., Inc. v. Atonio, 490 U.S. 642, 656-57 , 109 S.Ct. 2115, 2124-25 , 104 L.Ed.2d 733 (1989); Rose v. Wells Fargo & Co., 902 F.2d 1417, 1424 (9th Cir.1990) (citing Watson v. Fort Worth Bank & Trust, 487 U.S. 977, 994-96 , 108 S.Ct. 2777, 2788-90 , 101 L.Ed.2d 827 (1988)). It is not sufficient to present evidence raising an inference of discrimination on a disparate impact claim. The plaintiff “must actually prove the discriminatory impact at issue.” Rose, 902 F.2d at 1421 .

A prima facie case of disparate impact is “usually accomplished by statistical evidence showing ‘that an employment practice selects members of a protected class in a proportion smaller than their percentage in the pool of actual applicants.’ ” Robinson v. Adams, 847 F.2d 1315, 1318 (9th Cir.1988) (quoting Moore v. Hughes Helicopters, Inc., 708 F.2d 475, 482 (9th Cir.1983)). Although statistical data alone, in a proper case, may be adequate to prove causation, Wards Cove, 490 U.S at 650, 109 S.Ct. at 2121 , the “statistical disparities must be sufficiently substantial that they raise such an inference of causation.” Watson, 487 U.S. at 995 , 108 S.Ct. at 2789 ; see also Clady v. County of Los Angeles, 770 F.2d 1421, 1428-29 (9th Cir.1985).

IY

The district court found that no prima facie case of disparate impact was proven. Focusing on the final results of the promotion process, the district court noted that one out of six female applicants was promoted, whereas 3 out of 32 male applicants received a promotion to AIC. This meant that female applicants were promoted at a rate of more than 16 percent, compared to a promotion rate for male applicants of less than 10 percent.

We do not gainsay the district court’s reasoning as it pertains to the bottom line results of the promotion process. The problem, however, is that the promotion process included an intermediate stage that functioned as a pass or fail barrier to further consideration. Promotions to AIC were offered only to applicants who were interviewed by the selection committee, and interviews were granted only to those selected by the screening panel. It is at the intermediate screening stage that the postal inspectors direct their disparate impact claims. The nonadverse results of the ultimate promotion decisions cannot refute a prima facie case of disparate impact at the dis-positive interview selection stage. See Connecticut v. Teal, 457 U.S. 440, 452, 102 S.Ct. 2525, 2533 , 73 L.Ed.2d 130 (1982). Whether disparate impact was shown must address the results of the interview screening decisions, not simply the bottom line promotion decisions.

The postal inspectors urge us to adopt an even finer distinction. They contend that Teal requires analytical separation of the two screening rounds as well. No female applicant was selected to be interviewed in the first round. The postal inspectors wish to isolate this zero selection rate and make it the basis of their prima facie case. We see no sound reason to accept the distinction urged upon us. We separate the results of the interview selection stage from the results of the overall promotion process because the intermediate stage functioned as a pass or fail barrier to further consideration for promotion. No such barrier existed between the two screening rounds. Applicants who were not selected in the first round were again considered in the second round of interviews. Failure to be selected in the first round did not foreclose an applicant’s consideration for interview in the second round.

It is of no significance that the second round interviews were originally unplanned or limited to filling the San Francisco opening. The rationale, intent or motive underlying a challenged employment practice plays no part in a prima facie case of disparate impact. Watson, 487 U.S. at 988 , 108 S.Ct. at 2785 . The concern is the outcome of the practice at issue, not its underlying intent. See Griggs v. Duke Power Co., 401 U.S. 424, 432 , 91 S.Ct. 849, 854 , 28 L.Ed.2d 158 (1971) (holding that disparate impact claims are directed at “the consequences of employment practices, not simply the motivation”). The results of the first and second screening rounds represent the outcome of the same selection practice that the postal inspectors challenge. They cannot be analytically separated for purposes of disparate impact analysis.

We now consider the statistical conclusions that can be drawn from the results of both screening rounds. We observe initially that the probative value of any statistical comparison is limited by the small available sample. See Watson, 487 U.S. at 996-97 , 108 S.Ct. at 2790 (stating that statistical evidence may not be probative if it is based on a “small or incomplete data set”); Morita v. Southern Cal. Permanente Med. Group, 541 F.2d 217, 220 (9th Cir.1976) (“[SJtatistical evidence derived from an extremely small universe ... has little predictive value and must be disregarded.”) (quoting Harper v. Trans World Airlines, Inc., 525 F.2d 409, 412 (8th Cir.1975)). A sample involving 6 female applicants in a pool of 38 applicants is likely too small to produce statistically significant re-suits. Cf. Contreras v. City of Los Angeles, 656 F.2d 1267, 1272-73 (9th Cir.1981) (discounting probative value of statistical sample consisting of 57 test-takers, 17 of whom belonged in the plaintiffs’ protected class); Morita, 541 F.2d at 219-20 (finding 8-case sample too small). Assuming the data set here is adequately reliable, the evidence does not indicate a substantial statistical disparity.

The first step in a statistical analysis is to identify the base population for comparison. Generally, the appropriate population is the applicant pool or relevant labor market from which the positions at issue are filled. The composition of the applicant pool or relevant labor market is then compared to the composition of the successful applicants. See Wards Cove, 490 U.S. at 650-51 , 109 S.Ct. at 2121 ; Hazelwood Sch. Dist. v. United States, 433 U.S. 299, 308 , 97 S.Ct. 2736 , 53 L.Ed.2d 768 (1977) (holding appropriate comparison was between the racial composition of the qualified public school teachers in the relevant labor market and the racial composition of the school’s teaching staff). In the context of promotions, the appropriate comparison is between the composition of candidates seeking promotion and the composition of those actually promoted. See Waisome v. Port Authority, 948 F.2d 1370 , 1372 (2d Cir.1991).

Female applicants comprised 13.3 percent (2 of 15) of all those interviewed and 15.8 percent (6 of 38) of the original applicant pool. The percentage of interviewees who are female is nearly proportional to the percentage of applicants who are female. The 2.5 percent difference is not a substantial or significant statistical disparity.

As a “rule of thumb,” courts have also considered the so-called “four-fifths rule” suggested by the Equal Employment Opportunity Commission. See Clady, 770 F.2d at 1428 . Found in the Uniform Guidelines on Employee Selection Procedures, the four-fifths rule states that a selection practice is considered to have a disparate impact if it has a “selection rate for any race, sex, or ethnic group which is less than four-fifths {%) (or eighty percent) of the rate of the group with the highest rate.” 29 C.F.R. § 1607.4 (D) (2001). Applying this rule, we observe that the selection rate for female applicants to be interviewed was 33 percent (2 of 6) and the rate for male applicants was 41 percent (13 of 32). This means that the rate of selection for women was 81 percent of the rate of interview for men, again demonstrating that no disparate impact was shown.

V

The postal inspectors also falter in attacking a specific employment practice or selection criterion. They direct their disparate impact claims to “the decision-making process” or “the process by which the [screening] Panel evaluated applications.” They stress that no female applicants were initially screened for interviews. According to the postal inspectors, the imbalance in interview selections is attributable to the overall screening process.

Plaintiffs generally cannot attack an overall decisionmaking process in the disparate impact context, but must instead identify the particular element or practice within the process that causes an adverse impact. Wards Cove, 490 U.S. at 656-57 , 109 S.Ct. at 2124-25 . A decision-making process may be analyzed as a single employment practice if “the complaining party can demonstrate to the court that the elements of a respondent’s decisionmaking process are not capable of separation for analysis.” 42 U.S.C. § 2000e-2(k)(l)(B)(i).

The overall interview screening process included several discrete elements. Each member on the screening panel independently evaluated every applicant, measuring the applicant’s supervisor evaluations and the substantive contents of the application. The application tested a candidate’s proficiency on eleven validated competencies required for the AIC position. The panel members then met to compare their evaluations and discuss the applicants before collectively agreeing on the most qualified candidates.

The postal inspectors do not argue that the various elements of the screening process cannot be isolated for analysis. We doubt that the overall screening process should be treated as one employment practice for purposes of disparate impact analysis. It is not necessary to decide the issue because, in either event, the postal inspec tors fail to demonstrate the requisite causal connection between any selection practice and the claimed gender disparity.

The postal inspectors do not demonstrate that the screening process caused exclusion of female applicants from the interview stage because of the applicants’ gender. See Watson, 487 U.S. at 994 , 108 5.Ct. at 2789 (holding that plaintiff has burden to show that a particular employment practice “caused the exclusion of applicants for jobs or promotions because of their membership in a protected group”). In Griggs v. Duke Power Co., supra, the seminal disparate impact case, the Supreme Court recognized a disparate impact claim by black employees challenging the use of general aptitude tests and a requirement of a high school diploma for promotions. African Americans were underrepresented in the employer’s higher paying departments. The plaintiffs demonstrated a causal connection between the racial imbalance in the workforce and the challenged promotion practices by showing that black applicants scored disproportionately lower than white applicants on the tests and were less likely to have high school diplomas. Griggs, 401 U.S. at 431 n. 6, 91 S.Ct. at 853 n. 6.

In this case, by contrast, the postal inspectors fail to demonstrate how the screening process excludes female applicants due to gender. The elements of the screening process are facially gender neutral. An applicant is measured by performance on the validated competencies in the application and the strength of supervisor evaluations. There is no evidence that women perform worse than men on the competencies or that women receive poorer supervisor evaluations. Unlike in Griggs , where it was shown that the lower scores of black candidates were a factor in employment decisions that caused fewer black employees to be promoted, the postal inspectors here fail to show that women are measured lower than men on the neutral criteria involved in the screening decision. There is no evidence demonstrating that the neutral practices or criteria in the screening process “operate as ‘built-in head-winds’ ” for female applicants. Griggs, 401 U.S. at 432 , 91 S.Ct. at 854 . The postal inspectors fail to demonstrate the requisite showing of causation between a challenged employment practice and the claimed gender disparity.

VI

The postal inspectors fail to identify a specific employment practice that disproportionately excludes female applicants because of the applicants’ gender. The statistical evidence offered is of dubious reliance, and in any case, fails to show an actual disparity in the intermediate interview selection stage. The district court properly granted summary judgment in favor of the Service on the disparate impact claims.

AFFIRMED.

. The AIC vacancy in the Washington, D.C. office was eliminated as a result of restructuring, leaving four open positions. . If an interview was given to one more female applicant, there would be a higher percentage of female interviewees than female applicants. This potential reversal of fortunes underscores the problem in working with a small sample size. See Contreras, 656 F.2d at 1273 n. 4 ("Statistics are not trustworthy when minor numerical variations produce significant percentage fluctuations.”). . If we consider each interview screening process as a weighted, independent event and aggregate the results, the postal inspectors still fail to state a prima facie claim of discrimination. Under this lens, two-thirds (10) of the total interviewees (15) were selected from an initial applicant pool consisting of 38 applicants (6 women, 32 men). One-third (5) of the total interviewees (15) were selected from a second round applicant pool of 19 applicants (5 women, 14 men). Overall, women were still interviewed at 13.3 percent 0is), but women now make up 19.3 percent of the applicant pool. Eighteen percent (2/n) of women in the applicant pool were chosen for interviews, whereas men were chosen at a rate of 28 percent ('^ó). The selection rate for women is now 64 percent (I8/6s) that of men, and no longer within the 80 percent haven established by the EEOC Guidelines. Yet the apparent violation of the 80 percent rule causes little concern. A swing of one woman interviewed changes the female to male selection rate from 64 percent to 104.6 percent, and liability cannot turn on such statistical caprice.