I.
Former salmon cannery workers sued their employers for discrimination on the basis of race, advancing both disparate treatment and disparate impact claims under Title VII of the Civil Rights Act of 1964, 42 U.S.C. § 2000e et seq., and the Civil Rights Act of 1866, 42 U.S.C. § 1981. The district court declined to apply disparate impact analysis to certain subjective employment practices and this panel affirmed that decision. Atonio v. Wards Cove Packing Co., 768 F.2d 1120, 1132 & n. 6 (9th Cir.1985), withdrawn, 787 F.2d 462 (9th Cir.1985). An en banc panel decided that “disparate impact analysis may be applied to challenge subjective employment practices or criteria provided the plaintiffs have proved a causal connection between those practices and the demonstrated impact on members of a protected class.” Atonio, 810 F.2d 1477, 1482 (9th Cir.1987) (en banc). The en banc panel returned the cause to this panel to reconsider the district court’s disposition of the plaintiffs’ claims. Id. at 1486.
*442In our prior decisions we have presented the factual background of this case in considerable detail, and we will not repeat it here. See Atonio, 768 F.2d at 1122-24. We have also explained the legal principles governing analysis of Title VII disparate treatment claims. Id. at 1124-31. The en banc panel adopted the rule that disparate impact analysis may be applied to the “subjective” employment practices challenged in this case, but it did not explain in any detail how the analysis should be applied. See Atonio (en banc), 810 F.2d at 1482. We now provide that explanation, in light of the reasoning and rationale of the en banc panel in adopting impact analysis.
DISPARATE IMPACT ANALYSIS
A class claim of disparate impact is essentially an allegation that a disparity in the position of nonwhites and whites, often proved through statistical evidence, is “the systemic result of a specific employment practice that cannot be justified as necessary to the employer’s business.” Segar v. Smith, 738 F.2d 1249, 1267 (D.C.Cir. 1984), cert. denied, 471 U.S. 1115, 105 S.Ct. 2357, 86 L.Ed.2d 258 (1985). The quantity and quality of statistical evidence which will give rise to an inference that the disparity is caused by the employer’s practices is the same as that which will give rise to an inference of discriminatory intent. Id.
The crucial difference between a disparate treatment and a disparate impact allegation is the intermediate burden on the employer. To rebut the prima facie showing of disparate impact the employer may refute the statistical evidence as in the treatment claim and show that no disparity exists. But if the employer defends by explaining the reason for the disparity he must do more than articulate that reason. He must prove the job relatedness or business necessity of the practice. Albemarle Paper Co. v. Moody, 422 U.S. 405, 425, 95 S.Ct. 2362, 2375, 45 L.Ed.2d 280 (1975). The Supreme Court’s decision in Texas Dep’t of Community Affairs v. Burdine, 450 U.S. 248, 253, 101 S.Ct. 1089, 1093, 67 L.Ed.2d 207 (1981), that the burden of persuasion always stays with the plaintiff in a treatment case expressly preserved the different allocation of burdens in an impact case. The Court stated that it “recognized that the factual issues, and therefore the character of the evidence presented, differ when the plaintiff claims that a facially neutral employment policy has a discriminatory impact on protected classes.” Id. 450 U.S. at 252 n. 5, 101 S.Ct. at 1093 n. 5.
Precisely what the employer must prove will vary with the unique factors of different job settings, but “[t]he touchstone is business necessity.” Griggs v. Duke Power Co., 401 U.S. 424, 431, 91 S.Ct. 849, 853, 28 L.Ed.2d 158 (1971). Business necessity of employee selection criteria may be shown by demonstrating that the selection criteria applied are essential to job safety or efficiency, Dothard v. Rawlinson, 433 U.S. 321, 331 n. 14, 97 S.Ct. 2720, 2728 n. 14, 53 L.Ed.2d 786 (1977), or correlated with success on the job. Contreras v. City of Los Angeles, 656 F.2d 1267, 1280 (9th Cir.1981), cert. denied, 455 U.S. 1021, 102 S.Ct. 1719, 72 L.Ed.2d 140 (1982). In short, the employer must demonstrate the “manifest relationship” between the challenged practice and job performance. Griggs, 401 U.S. at 432, 91 S.Ct. at 854. Job relatedness is thus the means of proving “business necessity” when the purpose of a criterion is to predict the capacity of particular individuals to perform a job successfully-
When other employment practices are challenged, whose purpose is not to predict successful job performance, business necessity turns on proof of the burden or benefit to the business of the practice under scrutiny. See Schlei and Grossman, Employment Discrimination Law, 1329 (2d ed. 1983). Business necessity means more than a business purpose. Business necessity requires that a practice “must substantially promote the proficient operation of the business.” Chrisner v. Complete Auto Transit, Inc., 645 F.2d 1251, 1262 (6th Cir.1981). See also, Williams v. Colorado Springs School District No. 11, 641 F.2d 835, 842 (10th Cir.1981) (“The practice must be essential, the purpose compelling.”). Accord Crawford v. Western Electric Co., Inc., 745 F.2d 1373 (11th Cir.1984); *443Kirby v. Colony Furniture Co., 613 F.2d 696, 705 n. 6 (8th Cir.1980); Parson v. Kaiser Aluminum & Chemical Corp., 575 F.2d 1374, 1389 (5th Cir.1978), cert. denied, 441 U.S. 968, 99 S.Ct. 2417, 60 L.Ed.2d 1073 (1979); Head v. Timken Roller Bearing Co., 486 F.2d 870, 879 (6th Cir.1973); Robinson v. Lorillard Corp., 444 F.2d 791, 798 (4th Cir.), cert. dismissed, 404 U.S. 1006, 92 S.Ct. 573, 30 L.Ed.2d 655 (1971)
After the employer proves the business necessity of his practices, the plaintiff class has the opportunity to demonstrate that other employment practices or selection devices could serve the employer’s needs with a lesser impact on the protected class. Moody, 422 U.S. at 425, 95 S.Ct. at 2375; Moore v. Hughes Helicopters, Inc., 708 F.2d 475, 481 (9th Cir.1983); Chrisner, 645 F.2d at 1263. Whether the plaintiffs’ proposed alternative rebuts, or should prevail over, the employer’s proof of the business necessity of the original practice is then the ultimate determination to be made.
APPLICATION OF IMPACT ANALYSIS
A. Standard of Review
The ultimate finding of no discriminatory intent in a Title VII action is a factual finding that may be overturned on appeal only if it is clearly erroneous. Fed.R.Civ.P. 52(a); Anderson v. City of Bessemer, 470 U.S. 564, 573, 105 S.Ct. 1504, 1511, 84 L.Ed.2d 518 (1985); Pullman Standard v. Swint, 456 U.S. 273, 102 S.Ct. 1781, 72 L.Ed.2d 66 (1982); Gibbs v. Pierce County Law Enforcement Support Agency, 785 F.2d 1396, 1401 (9th Cir.1986). See also Kimbrough v. Secretary of the United States Air Force, 764 F.2d 1279, 1281 (9th Cir.1985) (“After a Title VII case is fully tried, we review the decision under the clearly erroneous standard applicable to factual determination.”). Under the clearly erroneous test, this court must affirm the district court’s determination unless “left with the definite and firm conviction that a mistake has been committed.” Gibbs, 785 F.2d at 1401 (quoting United States v. United States Gypsum Co., 333 U.S. 364, 395, 68 S.Ct. 525, 542, 92 L.Ed. 746 (1948)). The “ ‘district court must decide which party’s explanation of the employer’s motivation it believes.’ We will reverse that factual determination only if it is clearly erroneous ... and we will not ransack the record, searching for mistakes.” Casillas v. United States Navy, 735 F.2d 338, 342-343 (9th Cir.1984) (quoting United States Postal Service Bd. v. Aikens, 460 U.S. 711, 716, 103 S.Ct. 1478, 1482, 75 L.Ed.2d 403 (1983)).
Of course, we review legal questions de novo. United States v. McConney, 728 F.2d 1195, 1201 (9th Cir.) (en banc), cert. denied, 469 U.S. 824, 105 S.Ct. 101, 83 L.Ed.2d 46 (1984). The conclusion a district court reaches about whether a Title VII plaintiff has satisfied the elements of a prima facie case is reviewed de novo. See, e.g., Clady v. Los Angeles County, 770 F.2d 1421, 1427 (9th Cir.1985), cert. denied, — U.S.-, 106 S.Ct. 1516, 89 L.Ed.2d 915 (1986); Thorne v. City of El Segundo, 726 F.2d 459, 464 n. 5 (9th Cir.1983), cert. denied, 469 U.S. 979, 105 S.Ct. 380, 83 L.Ed.2d 315 (1984), appeal after remand, 802 F.2d 1131 (9th Cir.1986); Gay v. Waiters’ and Dairy Lunchmen’s Union, 694 F.2d 531, 540-45 & n. 13 (9th Cir.1982). We have also suggested, without deciding the question, that the appropriate standard for reviewing the lower court’s conclusion at the third stage of a discriminatory treatment case — proving that an employer’s proffered explanation for differential treatment is mere pretext — is also subject to de novo review. Thome, 726 F.2d at 465 & n. 6.
B. The Class Claims
As the en banc panel emphasized, a class action pattern and practice case is amenable to both treatment and impact analysis. Atonio, 810 F.2d at 1480. In reviewing the district court’s resolution of the class claims, our organizational principle is the practices complained of, rather than the mode of proof. But first we discuss the district court’s treatment of the statistical evidence offered by both parties.
*4441. Statistics
Statistical evidence is of critical value in creating an inference of either discriminatory intent or impact. We have recognized the importance of statistics as circumstantial evidence of discriminatory intent, but have cautioned that the weight given to them depends on “proper supportive facts and the absence of variables.” Spaulding v. University of Washington, 740 F.2d 686, 703 (9th Cir.), cert. denied, 469 U.S. 1036, 105 S.Ct. 511, 83 L.Ed.2d 401 (1984), overruled on other grounds, Atonio, 810 F.2d 1477 (en banc). The district court’s evaluation of conflicting statistics and determination of the probative weight they are to be accorded is a factual inquiry. Accordingly, we apply the clearly erroneous standard of review. Gay, 694 F.2d at 550; see also Allen v. Prince George’s County, Md., 737 F.2d 1299, 1303 (4th Cir.1984).
The plaintiffs introduced comparative statistics showing the disproportionate concentration of nonwhite persons in the lower paying jobs. In analyzing the evidence of disparate treatment, the district court began its inquiry by dividing the at-issue (non-cannery worker) jobs into two groups: unskilled and skilled.
Taking each group in turn, the court first found that the unskilled jobs were fungible, and, thus, comparative statistics were appropriate for use in establishing a prima facie case of discrimination. Since the comparative statistics showed a pattern of job segregation throughout the cannery work forces, the court found that the plaintiffs had established a prima facie case with respect to the unskilled jobs.
In considering the skilled positions, the district court had more difficulty in finding a prima facie case of intentional discrimination, because it did not consider plaintiffs’ statistical evidence probative. The court concluded that the practice of hiring through Local 37 had tended to distort the racial composition of the work force. Thus, when considering the skilled positions, the court found that statistics which merely highlight the segregation of whites and nonwhites between the at-issue and cannery worker jobs, without more, could not serve to raise an inference that the segregation is attributable to intentional discrimination against any particular race. Although we accept this finding, we stress that such statistics can serve to demonstrate the consequences of discriminatory practices which have already been independently established. Domingo v. New England Fish Co., 727 F.2d 1429, 1436 (9th Cir.) (per curiam), modified, 742 F.2d 520 (9th Cir.1984).
The cannery workers contend that the district court erred in failing to credit their comparative statistics when analyzing the skilled positions. The district court accorded these statistics, comparing the racial composition of the various job categories, little probative value because they did not reflect the number of employees possessing the requisite skills or those available for preseason work. This was error because when job qualifications are themselves at issue, the burden is on the employer to prove that there are no qualified minority people for the at-issue jobs. Kaplan v. International Alliance of Theatrical and Stage Employees, 525 F.2d 1354, 1358 n. 1 (9th Cir.1975); Wang v. Hoffman, 694 F.2d 1146, 1148 (9th Cir.1982). Furthermore, it is unrealistic to expect statistics to be calibrated to reflect preseason availability when the preseason starts only one month earlier than the season.
The comparative statistics offered by the cannery workers are sufficient to support an inference of discrimination in hiring practices both as to unskilled and skilled jobs. While the district court discounted the comparative statistics in evaluating the claim of intentional discrimination in skilled jobs we find them sufficiently probative of adverse impact. The statistics show only racial stratification by job category. This is sufficient to raise an inference that some practice or combination of practices has caused the distribution of employees by race and to place the burden on the employer to justify the business necessity of the practices identified by the plaintiffs. As the court stated in Domingo, comparative statistics demonstrate “the consequences *445of ... discriminatory hiring practices.” 727 F.2d at 1436.
Thus, in this case, because the comparative statistics support an inference of discriminatory impact, and because the cannery workers have identified certain practices which cause that impact, it is incumbent on the district court to evaluate the business necessity of the practices. Of course, it is also essential that the practices identified by the cannery workers be linked causally with the demonstrated adverse impact.
2. Employment Practices
a. Nepotism
The cannery workers contend that the district court erred in not giving more credit to their evidence of nepotism. The district court noted that “[rjelatives of whites and particularly (sic) nonwhites appear in high incidence at the canneries. However, defendants have established that the relatives hired in at-issue jobs were highly qualified for the positions in which they were hired and were chosen because of their qualifications.” The court also found that plaintiffs’ statistics failed to recognize that a number of persons became related through marriage after starting work at the canneries, and that the testimony showed “that numerous white persons who ‘knew’ someone were not hired due to inexperience, and whites hired were paid no more than nonwhites.” Therefore, the court concluded that there existed no “preference” for relatives at the canneries.
The district court subjected the cannery workers’ nepotism allegations to impact analysis, in accordance with Bonilla v. Oakland Scavenger Co., 697 F.2d 1297 (9th Cir.1982), cert. denied, 467 U.S. 1251, 104 S.Ct. 3533, 82 L.Ed.2d 838 (1984). We think the district court may have missed the point of Bonilla in evaluating nepotism at these canneries. If nepotism exists, it is by definition a practice of giving preference to relatives, and where those doing the hiring are predominantly white, the practice necessarily has an adverse impact on nonwhites. Id. at 1303. The evidence shows that of 349 nepotistic hires in four upper-level departments during 1970-75, 332 were of whites, 17 of nonwhites. That the court found individuals were hired for their skills and not because they were relatives serves to dispel the inference of discriminatory intent but it does not meet the defendants’ burden in refuting a claim of disparate impact. What is required is that the defendants prove the business necessity of the nepotism policy. Id.; Contreras, 656 F.2d at 1275-80. As we said in Bonilla, generally “nepotistic concerns cannot supersede the nation’s paramount goal of equal opportunity for all.” 697 F.2d at 1303.
b. Subjective Criteria
A crucial aspect of the cannery workers’ treatment claim was the alleged absence of job criteria and the latitude it allowed for subjective decision making. Courts recognize that subjective criteria are ready mechanisms for discrimination. See, e.g., EEOC v. Inland Marine Industries, 729 F.2d 1229, 1236 (9th Cir.), cert. denied, 469 U.S. 855, 105 S.Ct. 180, 83 L.Ed.2d 114 (1984); Domingo, 727 F.2d at 1436 n. 3. In evaluating a claim of dispar ate treatment, subjective criteria are suspect because they may mask the influence of impermissible racial bias in making hiring decisions. The district court considered the claim that there were no objective job criteria but found that there were in fact objective criteria. It adopted verbatim from the defendants’ pretrial order a list of qualifications which it found “reasonably required for successful performance” of a number of jobs. Opinion at 34. It did not, however, find that these specific criteria were actually applied by those who made hiring decisions, and at one point noted the “general lack of objective job qualifications.” Opinion at 60. The court said people were evaluated according to job-related criteria, but in context that statement apparently meant only that the general criteria of experience and skills were considered but subjectively evaluated by hiring officials. Thus the lists merely supported the conclusion that skill and/or experience were the general qualifications looked for *446in the hiring of employees for the specified jobs. The court also found that the necessary skills are not readily acquirable during the season, primarily due to the time restrictions involved, and that cannery worker jobs do not provide training for other positions. Further, the district court found that preseason availability is a necessary qualification for many of the positions, but that it is never a requirement for cannery worker jobs. While these findings are not clearly erroneous, and may serve to defeat the inference of discriminatory animus, they do not support a finding that there was no disparate impact occasioned by this practice.
The cannery workers allege that the lack of objective job qualifications and the consequent hiring on the basis of subjective evaluations has an adverse impact on nonwhites in the canning industry. The companies concede the causal relation between their hiring criteria and the number of nonwhites in the at-issue jobs, but argue that there are objective qualifications which differentiate among potential employees in such a way that there are no qualified nonwhites for the at-issue jobs. The district court, as discussed, found there were qualifications for the jobs, including specific skills and experience. We think the court must analyze whether these qualifications were actually applied in a non-discriminatory manner. The Supreme Court has held that only “non-discriminatory standards actually applied ” by employers are pertinent in a discrimination case.” Franks v. Bowman Transportation Co., 424 U.S. 747, 773 n. 32, 96 S.Ct. 1251, 1268 n. 32, 47 L.Ed.2d 444 (1976) (emphasis in original). There is anecdotal evidence which suggests that these criteria were not applied. For example, the district court found that reasonable qualifications for a dry tender engineer included “one year of related boat experience or six months engine mechanical experience and one season of tender experience.” But one dry tender engineer, who was a relative of a company home office employee, had had no mechanical experience or training other than performing preventive maintenance on his car, and no experience working on a boat.
More importantly, the court must bear in mind that where qualifications are at issue, the burden is on the employer to prove the lack of qualified people in the nonwhite group. Kaplan, 525 F.2d at 1358 n. 1. As we said in Wang, 694 F.2d at 1148, “[h]e cannot be required to prove that he qualified for promotions under a system he alleges to be discriminatory unless the legitimacy of the system is first established.” Finally, and most importantly, the court must make findings as to the job-relatedness of the criteria actually applied.
c. Separate Hiring Channels and Word-of-Mouth Recruitment
The cannery workers urge reversal on the ground that the district court’s findings failed to address the discriminatory nature of separate hiring channels and word-of-mouth recruitment. We are troubled by this omission. There is, however, sufficient indication that the court considered the practices and apparently found them explained by the companies’ professed concerns with honoring their commitments to various unions and finding appropriately skilled workers. See Nicholson v. Board of Education, 682 F.2d 858, 866 (9th Cir.1982).
The cannery workers argue that word-of-mouth recruitment and recruitment for skilled jobs in different channels from those used to fill unskilled jobs are a significant cause of the disparity in the jobs held by whites and nonwhites. Specifically, the companies sought cannery workers in Native villages and through dispatches from ILWU Local 37, thus securing a work force for the lowest paying jobs which was predominantly Alaska Native and Filipino. For other departments the companies relied on informal word-of-mouth recruitment by predominantly white superintendents and foremen, who recruited primarily white employees. That such practices can cause a discriminatory impact is obvious. See Domingo, 727 F.2d at 1435-36. This court has long recognized the contribution of separate hiring channels to proving the disparate impact of a pattern or practice of discrimination. In United States v. Iron-*447workers Local 86, 443 F.2d 544, 548 (9th Cir.), cert. denied, 404 U.S. 984, 92 S.Ct. 447, 30 L.Ed.2d 367 (1971), which involved both treatment and impact claims, we held that the “active recruitment of whites, while at the same time giving little or no publicity to information concerning procedures for gaining union membership, work referral opportunities, and the operation of the apprenticeship programs in the black community,” was probative of a pattern or practice of discrimination against blacks in the construction industry. Other courts, too, have long recognized that word-of-mouth recruiting “is discriminatory because of its tendency to perpetuate the all-white composition of a work force.” Barnett v. W.T. Grant Co., 518 F.2d 543, 549 (4th Cir.1975).
The defendant companies do not claim their practices have no impact, rather they assert business justifications for the practices. The companies say there are no people qualified for skilled jobs in the channels they tap for cannery worker positions, namely Local 37 and the Native villages. However, in considering the claims of the twenty-two individuals who alleged they had been discriminated against, the district court did not find they lacked qualifications, but rather that they did not make timely applications. Thus, there is evidence that some of the people counted in the comparative statistics may be qualified for skilled jobs, and it is not disputed they could fill the at-issue unskilled jobs.
We also point out that logic simply does not support the inference, in a time of widespread unemployment and underemployment, that persons who hold, or are willing to take unskilled jobs, lack the skills for other, more demanding and higher paying jobs. The burden must shift to the companies to prove the business necessity of this practice. The district court observed that it is not a reasonable business practice to seek skilled workers in remote, sparsely populated regions. We cannot agree without a more specific development of the facts and rationale that would explain why it would be unreasonable to notify all potential employees of all the job openings available.
We also agree with the plaintiffs that the district court may have erred in crediting the companies’ claims that the people in the channels from which it recruited for unskilled jobs were unavailable for preseason work and thus did not meet one of the requirements for many of the at-issue jobs. Residents of Alaska villages would logically be available for the preseason and the evidence simply does not support the broad conclusion that members of Local 37 were unavailable. The preseason begins in May and the season’s work begins in June and broad statistics do not tell us enough about the availability of otherwise qualified individuals.
d. Rehire Preferences
The salmon canneries give rehire preference to past employees in their old jobs. When jobs are racially stratified, giving rehire preference to former employees tends to perpetuate the existing stratification. International Brotherhood of Teamsters v. United States, 431 U.S. 324, 349, 97 S.Ct. 1843, 1861, 52 L.Ed.2d 396 (1977). It is not clear whether the district court considered whether this practice derived from an intent to discriminate. When it addressed the obvious disparate impact of the practice it held that rehires were justified by business necessity. The court found that the short season and dangers of the industry justified the rehire practice. This finding is supported by the evidence.
3. Race Labeling, Housing and Messing
Race labeling is pervasive at the salmon canneries, where “Filipinos” work with the “Iron Chink” before retiring to their “Flip bunkhouse.” The district court did not find the conduct laudatory but found that it was not “persuasive evidence of discriminatory intent.” Perhaps not, but the court must carry the analysis further and consider whether such a practice has any adverse impact upon minority people, i.e., whether it operates as a headwind to minority advancement.
*448The vast majority of cannery employees live at the canneries during the season in bunkhouses provided by the companies. The plaintiff class claimed that nonwhites, particularly Filipinos, were segregated from whites and placed in inferior bunkhouses because of racial discrimination. The district court found that the cannery workers established a prima facie case of intentional discrimination, but that the defendants’ evidence proved nondiscriminatory motivations which the class failed to prove pretextual. Specifically, the court found that the employees were housed by their time of arrival and by crew rather than with an intent to discriminate. The record contains sufficient evidence to support the district court’s conclusion that the companies articulated a nondiscriminatory reason for their practice.
Cannery workers were also fed separately from the remainder of the work force. They alleged that this was due to racial discrimination. The district court agreed that they had established a prima facie case of intentional discrimination, but that the defendants had met their burden of production and the cannery workers had not proved pretext. It is undisputed that the cannery worker mess halls served what is termed a “traditional” oriental menu. The district court noted that the Local 37 contract provided for a separate culinary crew, and that Filipino and Asian persons dominated the membership in Local 37. Further, the court found that the quality and quantity of food served in the respective mess halls were the responsibility of the respective cooks, and that the older cannery workers preferred the traditional menu, to which the younger workers acceded. The court concluded that complaints about the food were attributable to personal taste, and that the segregated messing arrangement was attributable to the union and not the conduct of defendants. There is support in the record for these findings, and we cannot find them clearly erroneous.
The district court also evaluated the complaints of segregated housing and messing under the impact theory and found that business necessity justified these practices. See Wambkeim v. J.C. Penney, 705 F.2d 1492 (9th Cir.1983) (impact analysis applies in employment benefits cases), cert. denied, 467 U.S. 1255, 104 S.Ct. 3544, 82 L.Ed.2d 848 (1984). The impact is clear in this case. The segregated housing aggravated the isolation of the non-white workers from the “web of information” spread by word-of-mouth among white people about the better paying jobs. See Domingo, 445 F.Supp. 421, 435 (W.D.Wash.1977), aff;d, 727 F.2d 1429 (9th Cir.), modified, 742 F.2d 520 (9th Cir.1984). But the district court found the companies could not be required to winterize all of their housing when bunkhouse assignment by- date of availability renders such an expenditure unnecessary. We hold that such a rationalization is not sufficient, without more, to sustain a finding of business necessity. Efforts to economize may be viewed as a business necessity only if the companies substantiate that these measures are clearly necessary to promote the proficient operation of the business. See Chrisner, 645 F.2d at 1262. Even if economizing is seen as a business necessity, the plaintiffs must have the opportunity to show that it could be accomplished with a lesser impact upon the nonwhite people in the cannery workforce.
The court found the separate mess facilities mandated by the employer-union agreements with Local 37. Since it also correctly noted that an agreement with a union will not immunize an employer from discrimination claims, Williams v. Owens-Illinois, Inc., 665 F.2d 918, 926 (9th Cir.), cert. denied, 459 U.S. 971, 103 S.Ct. 302, 74 L.Ed.2d 283 (1982), we are unsure what its conclusion was as to the discriminatory impact of separate messing.
In assessing how racial labeling and segregated housing and messing facilities may cause an adverse impact we suggest that the court consider the message that such practice conveys to the general population. As the Supreme Court has warned:
The [“whites only”] message can be communicated to potential applicants more subtly but just as clearly by an employer’s actual practices — by his consistent discriminatory treatment of actu*449al applicants, by the manner in which he publicizes vacancies, his recruitment techniques, his responses to casual or tentative inquiries, and even by the racial or ethnic composition of that part of his work force from which he has discriminatorily excluded members of minority groups.
Teamsters, 431 U.S. at 365, 97 S.Ct. at 1870.
C. Individual Claims
Twenty two plaintiffs alleged that they were either overtly discriminated against in the hiring for at-issue positions, or that they were deterred from seeking at-issue positions because of the defendants’ alleged history of pervasive discrimination. The district court correctly noted that a plaintiff seeking relief under 42 U.S.C. § 1981 must show intentional discrimination and then analyzed the § 1981 and the Title VII treatment claims under the McDonnell Douglas test. McDonnell Douglas Corp. v. Green, 411 U.S. 792, 802, 93 S.Ct. 1817, 1824, 36 L.Ed.2d 668 (1973). The court found the individuals had failed to establish a prima facie case because they could not show they had applied for existing job openings and thus no inference of discriminatory intent arose. They had made oral inquiries, which were not considered applications, or their applications were untimely. Applications could be untimely if made too early or too late. Testimony showed that some plaintiffs had orally inquired during one season about positions for the next season a year away, and such inquiries were not considered an application unless followed up by a written application to the home office during the winter. Conversely, because the companies generally received far more applications than there were job vacancies, an application was untimely if received after the opening was filled. The district court found that the defendants did not treat whites and nonwhites differently in these respects. The court also found that some applicants had been unavailable for preseason work and, therefore, unavailable for the positions they desired. While there is evidence in the record to support the district court s findings regarding these individual claims, the findings are premature in light of the decision that the practices of these employers must be evaluated for disparate impact.
The cannery workers argue persuasively that the companies’ use of separate hiring channels and word-of-mouth recruitment, and their failure to announce vacancies should serve to excuse the cannery workers from the necessity of establishing the timeliness of their applications and automatically elevate oral inquiries to the status of applications. See O’Brien v. Sky Chefs, 670 F.2d 864, 868 (9th Cir.1982), overruled on other grounds, Atonio, 810 F.2d 1477 (en banc). The cannery workers’ argument derives from a discussion of damages issues in Domingo, 727 F.2d at 1445. In Domingo we said it would be an unrealistic burden on claimants to prove timely applications when application procedures were informal and word-of-mouth recruitment made it difficult for present or prospective employees to become aware of openings when they occurred. Id. For the same reasons, if the district court in this case finds that the challenged practices violate Title VII under the impact analysis, it must then conduct additional proceedings to determine appropriate individual relief, even though individuals have not persuaded the court of their disparate treatment. See Teamsters, 431 U.S. at 361, 97 S.Ct. at 1867; Franks, 424 U.S. at 773 n. 32, 96 S.Ct. at 1268 n. 32.
D. The Motion for Attorney’s Fees
We decline to entertain any motion for attorney’s fees at this point in this litigation. There are issues of fact and law remaining for determination. We leave to the district court to determine, upon proper motions, properly supported, whether and to what extent any party is a prevailing party for the purposes of an award of attorney’s fees. See Hensley v. Echerhart, 461 U.S. 424, 103 S.Ct. 1933, 76 L.Ed.2d 40 (1983). If any fee award is made it shall include appropriate consideration of fees for this appeal and all proceedings in the district court.
*450The judgment is VACATED and the cause is REMANDED for further proceedings consistent with this opinion.