with whom FLAUM, Circuit Judge, joins, dissenting.
Whether employers should restrain adults from engaging in acts hazardous to their children is an ethical, medical, economic, and political problem of great complexity. But this is a statutory case, and we must implement the law rather than give our own answer. Johnson’s policy is sex discrimination, forbidden unless sex is a “bona fide occupational qualification”— which it is not.
I
Title VII of the Civil Rights Act of 1964 forbids employers “to discriminate against any individual ... because of such individual’s ... sex”, 42 U.S.C. § 2000e-2(a)(l), unless sex is a “bona fide occupational qualification reasonably necessary to the normal operation of that particular business or enterprise”, 42 U.S.C. § 2000e-2(e)(l). Both the district court and the majority believe that the fetal protection policy may be lawful despite the absence of a BFOQ.
A
Johnson uses sex as a ground of decision. The fetal protection policy applies to all women and no men. It is not written without reference to gender, having an un*909welcome side effect. Cf. Personnel Administrator of Massachusetts v. Feeney, 442 U.S. 256, 99 S.Ct. 2282, 60 L.Ed.2d 870 (1979). Differences between the sexes are its stated rationale. Only women transmit lead to children during pregnancy. Because a few women become pregnant with elevated levels of lead in the blood (in four years, eight out of an unknown number), Johnson excludes all women from the danger zone. This treats an employee not as an individual but as a woman. A plan using sex as a criterion and justified by arguments referring to sex is “discrimi-nat[ion] ... because of ... sex”. Los Angeles Department of Water & Power v. Manhart, 435 U.S. 702, 98 S.Ct. 1370, 55 L.Ed.2d 657 (1978).
General Electric Co. v. Gilbert, 429 U.S. 125, 97 S.Ct. 401, 50 L.Ed.2d 343 (1976), held that a rule distinguishing on account of pregnancy is not sex discrimination, because women are in both the “pregnant” and “non-pregnant” groups. See also Geduldig v. Aiello, 417 U.S. 484, 94 S.Ct. 2485, 41 L.Ed.2d 256 (1974). The Court saw the line as one between pregnant employees and all others, a line based on something other than sex (or at least something in addition to sex). Johnson’s line based on ability to become pregnant, however, is assuredly based on sex. That would be ground for distinguishing Gilbert, but Congress interred Gilbert in 1978 by enacting the Pregnancy Discrimination Act, 42 U.S.C. § 2000e(k) (the PDA), which provides:
The terms “because of sex” or “on the basis of sex” [in Title VII] include, but are not limited to, because of or on the basis of pregnancy, childbirth, or related medical conditions; and women affected by pregnancy, childbirth, or related medical conditions shall be treated the same for all employment-related purposes ... as other persons not so affected but similar in their ability or inability to work....
This amendment to Title VII makes distinctions based on women’s ability to bear children sex discrimination. It also has a built-in BFOQ standard: unless pregnant employees differ from others “in their ability or inability to work”, they must be treated “the same” as other employees “for all employment-related purposes”. Although located in a definitional provision, the language after the semicolon is substantive and governs Johnson’s plan.
Wright v. Olin Corp., 697 F.2d 1172 (4th Cir.1982), the only other appellate decision that has dealt with a fetal protection policy similar to Johnson’s, took a different view.1 Wright observed that a policy using sex as a ground of decision may cause women no more injury than a policy neutral with regard to sex, yet having a disparate impact. A policy designed to promote the health of offspring of both sexes is neutral in objective. A sex-neutral policy is judged under an approach more lenient than the BFOQ standard. Believing that a fetal protection policy rests on strong justifications, Wright treated the policy as sex-neutral so that it could sustain a rule functionally identical to Johnson’s. 697 F.2d at 1184-92.
This makes things turn not on whether the employer uses sex as a ground of decision but on whether the employer uses sex to serve a “good” policy. If the policy is beneficent and the injury to women “tolerable” in light of the interests served, the court changes the standard of inquiry. Yet whether a policy is “good” is a statutory question, governed by the BFOQ test stated in § 2000e-2(e)(l) and the supplemental rule of the PDA that women and men who are “similar in their ability or inability to work” must be treated the same. A court’s belief that a good end is in view does not justify departure from the statutory framework; it is an occasion for applying the statutory framework. Wright ig*910nored the PDA and inverted ordinary rules of statutory interpretation when stating (with echoes in the majority’s opinion today): “The inappropriateness of applying the overt discrimination/b.f.o.q. theory of claim and defense ... is that, properly applied, it would prevent the employer from asserting a justification defense which under developed Title VII [disparate impact] doctrine it is entitled to present.” 697 F.2d at 1182 n. 21. In other words, this must be a disparate impact case because an employer couldn’t win it as a disparate treatment case. If the rigors of the BFOQ suggest the need for a fresh approach, that is a job for another branch. See TV A v. Hill, 437 U.S. 153, 98 S.Ct. 2279, 57 L.Ed.2d 117 (1978). Cf. Richmond v. J.A. Croson Co., — U.S. -, 109 S.Ct. 706, 720-23, 102 L.Ed.2d 854 (1989) (the standard for reviewing decisions based on race does not depend on the strength of the justification or on who gains from its use).
In principle a court could make the legal standard turn on what the authors of a rule are trying to accomplish, rather than on the criteria they use to get there. Whether to do so was the nub of the debate in Man-kart. Los Angeles adopted a pension plan that collected more per month from women during employment and paid retired women the same per month as men. The city defended the policy by observing that the sums collected matched the actuarial value of the payments over the retired employr ees’ lives, because the average woman lives longer than the average man and so receives more monthly checks. Was it sex discrimination prohibited by Title VII? The women said yes, on the ground that every woman paid more per month than every man. The difference was based on sex and justified by an effect (longevity) linked with sex. The employer said no, on the ground that the pension was worth the contributions ex ante to men and women equally, and the groups therefore were treated equally. Only by using sex as a ground of decision, the employer pointed out, could it achieve actuarial equality. The Court held that the policy was sex discrimination, because the city used sex as the basis of decision. That this criterion produced equal outcomes for groups was irrelevant in the Court’s view, because Title VII requires employees to be treated as individuals. To say that sex had been considered in order to achieve equal group averages, the Court believed, was to confess a violation of the law. See also Arizona Governing Committee v. Norris, 463 U.S. 1073, 103 S.Ct. 3492, 77 L.Ed.2d 1236 (1983).
Manhart establishes two propositions that together are fatal to Johnson’s position. First, Title VII requires equal treatment of employees as persons rather than equal treatment of groups defined by sex (or race, or any other criterion listed in the statute). Observing that the average member of one group does as well as the average member of another does not support the use of any given ground of decision; indeed, resorting to notions of group equality begs the question how a statute presumptively forbidding the use of these criteria could permit them to be used to justify conduct. Second, that equal treatment of employees as persons will lead to higher costs of employing persons of a given sex is no defense. An obligation to pay men and women equal amounts per month after retirement and deduct from pay the same amount per month during employment means that the employer must contribute greater sums per month during women's working years. That incremental cost of female employees is, as the Court construed the statute, no reason to treat women differently. Again this is part of the idea of an anti-discrimination law. Women may have higher pension costs, or higher medical insurance costs (because of pregnancy), or take more days off because of sickness, or have shorter careers (again because of children). Title VII excludes these as reasons for preferring men, e.g., Orzel v. City of Wauwatosa, 697 F.2d 743, 755 (7th Cir.1983). The PDA, requiring equal treatment of employees “similar in their ability or inability to work”, reinforces this conclusion.
Manhart’s approach is the norm in anti-discrimination law. Consider, for example, the Supreme Court’s first important sex *911discrimination case under Title VII, Phillips v. Martin Marietta Corp., 400 U.S. 542, 91 S.Ct. 496, 27 L.Ed.2d 613 (1971). The employer refused to hire women with preschool-age children. Such a rule might have had benefits for the children, neutral with respect to sex, but the policy was not neutral. The Court brusquely held that the policy is disparate treatment, unlawful unless supported by a BFOQ. Next consider the “bottom line” defense rejected in Connecticut v. Teal, 457 U.S. 440, 102 S.Ct. 2525, 73 L.Ed.2d 130 (1982). The employer contended that so long as it hired persons in a mixture reflecting the applicants or population (the “bottom line”), it did not matter that steps along the way appeared to filter out members of one or another group. The Court disagreed, because an argument based on the bottom line treats persons as members of a group defined by sex (race, etc.) and proceeds as if the groups (rather than the applicants and employees) held the right to equal treatment. Title VII, as the Court saw it, requires the employer to treat the applicant as a person without regard to race, sex, etc., on every occasion. These eases, and others like them, dispatch arguments of the sort that because the employer pursues an end, children’s welfare, defined without regard to sex, a rule of decision that uses sex as a criterion should be treated as if sex-neutral. Cf. Palmore v. Sidoti, 466 U.S. 429, 104 S.Ct. 1879, 80 L.Ed.2d 421 (1984) (same approach in constitutional law).
When the employer engages in sex, race, or age discrimination in an effort to protect customers or members of the public, courts regularly see this as disparate treatment, for which a BFOQ is essential. See Johnson v. Mayor and City Council of Baltimore, 472 U.S. 353, 105 S.Ct. 2717, 86 L.Ed.2d 286 (1985); Western Air Lines, Inc. v. Criswell, 472 U.S. 400, 422-23, 105 S.Ct. 2743, 2755-56, 86 L.Ed.2d 321 (1985); Hahn v. City of Buffalo, 770 F.2d 12 (2d Cir.1985); Heiar v. Crawford County, 746 F.2d 1190 (7th Cir.1984). There is no reason why things should be different when fetuses, rather than adult bystanders, are the object of the employer’s protection.
Furnco Construction Corp. v. Waters, 438 U.S. 567, 576, 98 S.Ct. 2943, 2949, 57 L.Ed.2d 957 (1978), says that the details of the “prima facie case” of discrimination, and the order of proof and rebuttal at trial, are flexible and should be adjusted to the circumstances. See also Price Waterhouse v. Hopkins, — U.S.-, 109 S.Ct. 1775, 104 L.Ed.2d 268 (1989). Our case does not concern the order of proof and methods of inference; Johnson expressly uses sex to make decisions. When Wisconsin excluded male guards from a women’s prison, we saw this as disparate treatment and searched for a BFOQ. Torres v. Wisconsin Department of Health and Social Services, 859 F.2d 1523, 1526-28 (7th Cir.1988) (en banc). Here, too, there is disparate treatment. Fetal protection policies therefore may be justified, if at all, as BFOQs.
A word about the enforcement policy of the Equal Employment Opportunity Commission. The Commission told its staff to use Wright and the elaboration of its approach in Hayes v. Shelby Memorial Hospital, 726 F.2d 1543 (11th Cir.1984), when investigating fetal protection policies. Ordinarily the EEOC’s views about the meaning of Title VII are entitled to deference. This policy, however, does not so much interpret Title VII as adopt guidelines for prosecution. Prosecutorial guidelines reflect limitations on the agency’s resources and existing judicial interpretations; they do not define the meaning of the law. The Commission told its staff to follow the courts.2 The guidelines do not discuss the PDA or Manhart and give no reason for adopting the Wright-Hayes approach as opposed to the BFOQ standard. This may be sound from a prosecutorial perspective but does not address the question in our case. If the EEOC’s statement is designed as an interpretive rule, it is neither rea*912soned nor consistent with Title VII. See Public Employees Retirement System of Ohio v. Betts, — U.S.-, 109 S.Ct. 2854, 2863, 106 L.Ed.2d 134 (1989).
B
The statute allows an employer to show that consideration of sex is “reasonably necessary to the normal operation of that particular business”. Dothard v. Rawlinson, 433 U.S. 321, 332-37, 97 S.Ct. 2720, 2728-30, 53 L.Ed.2d 786 (1977); Criswell, 472 U.S. at 422-23, 105 S.Ct. at 2755-56; Torres, 859 F.2d at 1527. The plaintiffs argue that “the sex based practice involved here should ... be held invalid” (Br. 44) because the reasons Johnson offers cannot establish a BFOQ even if factually supported. The Fourth Circuit (unlike the majority of this court) believes that as a matter of law a fetal protection policy does not satisfy the standards for a BFOQ, see Wright, 697 F.2d at 1185 n. 21, and I think it has this much correct for two reasons: Johnson’s stated objectives are insufficient, and even if sufficient do not apply to all women.
Johnson defends its fetal protection policy on the basis of concern for the welfare of the next generation, an objective unrelated to its ability to make batteries (§ 2000e-2(e)(l) speaks of the “operation of the business”) or to any woman’s “ability or inability to work” (the standard of the PDA). Johnson allowed women to work until 1982, without ill effects on its business; for all we know (the record is silent), other firms in the same business employ women in the kinds of jobs from which Johnson excludes them. The majority does not mention the PDA, which, added to the BFOQ rule, puts out of bounds the justifications Johnson offers.
At oral argument before the panel counsel offered a new defense of Johnson’s policy: that it is morally required to protect children from their parents’ mistakes. This justification is redolent of Muller v. Oregon, 208 U.S. 412, 28 S.Ct. 324, 52 L.Ed. 551 (1908), which sustained a statute curtailing women’s hours of work on the ground that maternal functions unsuited women for long hours. The Court wrote:
[B]y abundant testimony of the medical fraternity continuance for a long time on her feet at work, repeating this from day to day, tends to injurious effects upon the body, and as healthy mothers are essential to vigorous offspring, the physical well-being of woman becomes an object of public interest and care in order to preserve the strength and vigor of the race.... [H]er physical structure and a proper discharge of her maternal functions — having in view not merely her own health, but the well-being of the race — justify legislation to protect her from the greed as well as the passion of man. The limitations which this statute places upon her contractual powers ... are not imposed solely for her benefit, but also largely for the benefit of all.
208 U.S. at 421-22, 28 S.Ct. at 326-27. The “abundant testimony of the medical fraternity” turned out to be the triumph of imagination over data. Dangers decried in Muller are today perceived as chimerical, excuses for blockading women as effective competitors of men in the labor force. Legislation of the sort allowed by Muller “protected” women out of their jobs by making women less attractive as employees. An employer that needed flexibility in assigning hours of work had to hire men; women were consigned to jobs with regular hours but lower wages. See Elisabeth M. Landes, The Effect of State Maximum Hours Laws on the Employment of Women in 1920, 88 J.Pol.Econ. 476 (1980) (finding that “protective” legislation reduced women’s hours, hourly wages, and annual income). Such laws also treat women in a stereotypical way. State laws requiring or allowing employers to treat women differently, on the assumption that women are less able than men to take the precautions essential for healthy children, are preempted by Title VII — not because of an express preemption clause, but because no state law may require or excuse a violation of federal law. Rosenfeld v. Southern Pacific Co., 444 F.2d 1219, 1225-27 (9th Cir. 1971). Statutes of the sort sustained in Muller, supported by the justifications ad*913vanced in Muller, are museum pieces, reminders of wrong turns in the law. It is not enough to say that Johnson is a private employer while Muller dealt with state laws. Title VII is addressed to private employers. The question is whether a justification of a particular kind is an acceptable defense of sex discrimination. This justification is not. No legal or ethical principle compels or allows Johnson to assume that women are less able than men to make intelligent decisions about the welfare of the next generation, that the interests of the next generation always trump the interests of living woman, and that the only acceptable level of risk is zero. “[T]he purpose of Title VII is to allow the individual woman to make that choice for herself.” Dotkard, 433 U.S. at 335, 97 S.Ct. at 2730.
Although some women may become pregnant, and a subset of their children might suffer, Johnson cannot exclude all fertile women from its labor force on their account. Most women in an industrial labor force do not become pregnant;3 most of these will have blood lead levels under 30 ¡ug/dl (only about lh of the employees exposed to lead at Johnson’s plants have higher levels); most of those who become pregnant with levels exceeding 30 ¡ug/dl will bear normal children (Johnson reports no birth defects or other abnormalities in the eight pregnancies among its employees).4 Concerns about a tiny minority of women cannot set the standard by which all are judged. An employer establishes a BFOQ only if there is “a factual basis for believing that all or substantially all women would be unable to perform safely and efficiently the duties of the job involved.” Weeks v. Southern Bell Telephone & Telegraph Co., 408 F.2d 228, 235 (5th Cir.1969), quoted with approval in Dothard, 433 U.S. at 333, 97 S.Ct. at 2728, and adopted as the law of this circuit in Torres, 859 F.2d at 1527, 1530. Fear of prenatal injury (which has not happened in the history of the employer) is a far cry from something that prevents “all or substantially all” women from doing their jobs.
To meet the “all or almost all” requirement, Johnson relies on an elaboration of the BFOQ for age discrimination offered in customer-safety cases: if all persons over a certain age are unsafe pilots (drivers, police officers, etc.), or if it is impossible to tell which of the older employees is unsafe, then age may be a BFOQ for employment. E.g., EEOC v. Mississippi State Tax Commission, 873 F.2d 97, 98 (5th Cir.1989) (en banc). The genesis of the “impossible to tell” branch of the BFOQ in age cases is that an unsafe employee can’t do the job the employer demands. It is word play to say that “the job” at Johnson is to make batteries without risk to fetuses in the same way “the job” at Western Air Lines is to fly planes without crashing. When an employer genuinely can’t tell which employees are unfit, and when unfitness for work is a serious problem, it may use a proxy that otherwise would be off limits. Johnson is not using sex to avert harm to customers. To repeat the statutory language: sex is a BFOQ when its use is “reasonably necessary to the normal operation of [the employer’s] particular business” (the BFOQ rule), and employers must treat men and pregnant women equally with respect to their “ability or inability to work” (the PDA). Risk to fetuses falls outside these rules.
Johnson might be concerned about cost. It could have argued that the only alternative to the fetal protection policy is a much cleaner workplace, so that no employee’s *914blood lead level exceeds 30 ¡ug/dl, which would be prohibitively expensive and lead it to close the business. Johnson does not make that argument, so the majority properly does not decide whether it would be a BFOQ. Another potential cost comes from tort law. Perhaps Johnson anticipated litigation filed by children injured by lead. The firm does not make this argument either, and so far as I know no child has recovered a judgment on account of parents’ occupational exposure to lead. Security National Bank v. Chloride, Inc., 602 F.Supp. 294 (D.Kan.1985), the only reported case, ended in a jury verdict for the employer even though the child argued that the employer violated OSHA’s maximum exposure rules. (Johnson says that it complies with OSHA’s rules.) Anyway, the prospect of tort judgments means only that female employees’ average cost to Johnson exceeds that of male employees. Title VII requires employers to deal with individual employees rather than with group averages. No firm could exclude women from its work force by saying that higher costs of pensions and health care made them too costly.5 If these costs do not establish a BFOQ, could not establish it even in principle, how may the prospect of tort judgments do so? Title VII applies even when — especially when — discrimination is rational as the employer sees things.
All of this is not to say where wisdom lies. An employer is better situated than its workers to gather and interpret scientific data, for medical studies are hard to evaluate, and the need to earn a living may induce employees to give too little weight to the interests of their offspring. A recent poll showing that more than 20% of American adults do not know that the Earth orbits the Sun does not lend confidence in the ability of the populace to make decisions that depend on scientific or medical data.6 If any given employer errs, the forces of competition open opportunities elsewhere. No one can treat lightly the possibility of injury to future children, who cannot protect themselves or participate in the decisions that will govern their lives. Trying to find the “right” accommodation would rob many a person of sleep — for rigorous implementation of fetal protection policies could close more than 20 million jobs to women,7 while failure to do any*915thing causes injury to unknown numbers. Under the PDA neither the employer nor the court is authorized to essay an answer to this social puzzle. The disparate treatment — BFOQ approach governs, and it resolves today’s dispute. Title VII gives parents the power to make occupational decisions affecting their families. A legislative forum is available to those who believe that such decisions should be made elsewhere.8
II
Having adopted the Wright-Hayes approach, we still should not affirm the district court’s judgment. Hayes opined that a fetal protection policy applicable only to women violates Title VII
unless the employer shows (1) that a substantial risk of harm exists and (2) that the risk is borne only by members of one sex; and (3) the employee fails to show that there are acceptable alternative policies that would have a lesser impact on the affected sex.
726 F.2d at 1554. At the time Wright and Hayes were decided, and when the EEOC issued its policy statement, courts believed that “business necessity” in a disparate impact case is a defense. “Business necessity” and “BFOQ” were not so distinct. We know from Wards Cove Packing Co. v. Atonio, — U.S.-, 109 S.Ct. 2115, 104 L.Ed.2d 733 (1989), however, that the plaintiff bears the burden of persuasion on all questions in every disparate impact case, as the majority today emphasizes. So the Wright-Hayes standard has been watered down. The court’s “adoption” of Wright, Hayes, and the EEOC’s policy statement is thus in practice more favorable to employers than the Fourth and Eleventh Circuits (and the EEOC) anticipated their approach would be. The plaintiff won in Hayes; she would lose under the majority’s approach.
Even on the majority’s un-demanding standard, however, there are material disputes. The EEOC’s policy statement properly criticized the district court for granting summary judgment despite evidentiary disputes material to the application of the Wright-Hayes standard. Notice n915.034 at 9 n. 22.
Substantial Risk. Is there a substantial risk of harm to the offspring of female employees? That lead in the blood is dangerous no one doubts, although experts dispute how much is too much and whether lead is more risky to the fetus and developing infants than to adults.9 The extent to which lead in a mother’s blood, at levels to which Johnson exposes its employees, endangers the fetus is a subject on which there is additional dispute. Showing great injury at 100 gg/dl is one thing; showing risks when no one has levels over 50 jug/dl (and, per OSHA’s rules, anyone wanting to get below 30 gg/dl is entitled to a respirator) is another. The record does not quantify the risks at the levels OSHA permits. It also does not reveal the extent to which lead crosses the placenta. Johnson’s chief medical consultant, Dr. Fishburn, believes that risk is greatest in the first weeks of pregnancy, before women can withdraw to lead-free environments. Experts take the contrary view that lead does not cross the placenta until late in pregnancy,10 and that because lead levels in the blood fall sub*916stantially within 12 weeks of the last exposure,11 the danger is reduced if women are removed from contact with lead promptly on becoming pregnant.
It is painful to see conflicts of this kind settled by litigation; judges cannot unravel medical mysteries by observing scientists’ demeanor on the stand. This is not how scientists resolve disputes among themselves. Demeanor tells the judge only whether the scientist believes what he says, something almost irrelevant to a scientific dispute. Scientists formulate hypotheses, collect data, and apply statistical methods to assess them; judges and jurors find this process alien. Yet so long as the substantive rule of law requires a court to resolve scientific disagreements — which the Wright-Hayes standard does, though the BFOQ standard avoids the problem — the judge must follow the rules, which means that material disputes must be resolved at trial.
A small risk, even if compellingly documented, is not enough to exclude women from employment. How great is too great? Most women do not become pregnant in any given year, and most female employees do not have blood lead levels exceeding 30 fig/dl, so average risk to the employee population may be small. If a woman becomes pregnant with a blood lead level of 40 /xg/dl, is the risk one learning disability in two pregnancies? One in two thousand? One in two million? These figures imply different policies, yet we do not know which is correct. How risky is a blood lead level exceeding 30 jug/dl compared with other hazards? Most comparisons show that smoking and drinking are quite dangerous to fetuses, more so than many contaminants found in the workplace.12 Some 2,500 children under the age of one die every year because of their parents’ smoking; others never make it to birth.13 The hazards created by occupational chemicals span many orders of magnitude: some are safer than the sweeteners we wolf down, some are dangerous indeed.14 Where does lead fit on that spectrum? I cannot believe that Johnson would *917be entitled to fire female employees who smoke or drink during pregnancy — let alone to fire all female employees because some might smoke or drink — which makes it hard to exclude women to curtail risk from other substances.
How does the risk attributable to lead compare, say, to the risk to the next generation created by driving a taxi? A female bus or taxi driver is exposed to noxious fumes and the risk of accidents, all hazardous to a child she carries. Would it follow that taxi and bus companies can decline to hire women? That an employer could forbid pregnant employees to drive cars, because of the risk accidents pose to fetuses? For all we can tell, accepting Johnson’s argument compels us to answer “yes” to these questions, which simply points up the need to quantify the hazard. It also points up the political nature of the dispute. How much risk is too much is a moral or economic or political question, one ill suited to the processes of litigation and not the sort of question Title VII puts to a judge.15
The most concerted effort to estimate the risks lead poses to offspring was conducted by OSHA in the course of promulgating its lead rules. OSHA considered and rejected a proposal to exclude women capable of bearing children from jobs in which blood lead levels may exceed 30 jug/100 g. It stated:
Industry testimony further suggests that women of childbearing potential could be “protected” by excluding them from employment in many parts of the lead in-dustry_ The record in this rulemak-ing is clear that male workers may be adversely effected [sic] by lead as well as women. Male workers may be rendered infertile or impotent, and both men and women are subject to genetic damage which may affect both the course and outcome of pregnancy. Given the data in this record, OSHA believes there is no basis whatsoever for the claim that women of childbearing age should be excluded from the workplace in order to protect the fetus or the course of pregnancy. Effective compliance with all aspects of these standards will minimize risk to all persons and should therefore insure equal employment for both men and women.
43 Fed.Reg. 52953, 52966 (1978). OSHA’s regulations require employers to make respirators available, so that both men and women desiring children can reduce their blood lead levels. 29 C.F.R. § 1910.1025(f)(l)(iii). They also require employers to conduct annual educational programs “with particular attention to the adverse reproductive effects on both males and females”, § 1910.1025(( )(l)(v)(D). In order to say that the risk to offspring at firms complying with OSHA’s rules (as Johnson says it does) is so “substantial” that a woman should not be allowed to work at any job where there is the slightest chance of a blood lead level exceeding 30 ¡ig/dl, a court must disagree with the judgment of OSHA that the 50 /xg/100 g limit, plus the availability of respirators to employees seeking to attain a level of 30 /xg/100 g, is enough. My colleagues essentially take judicial notice that OSHA is wrong, majority op. 899-900, an extraordinary step. The record does not contain evidence sufficient to contradict OSHA’s conclusion, let alone evidence so lopsided that summary judgment is appropriate.
One more observation. “Substantial risk” must mean substantial net risk. Ex-*918eluding women from industrial jobs at Johnson may reduce risk attributable to lead at the cost of increasing other hazards. There is a strong correlation between the health of the infant and prenatal medical care; there is also a powerful link between the parents’ income and infants’ health, for higher income means better nutrition, among other things. See Aaron Wildavsky, Searching for Safety 59-72 (1988); Victor R. Fuchs, How We Live 31-40 (1983). Removing women from well-paying jobs (and the attendant health insurance), or denying women access to these jobs, may reduce the risk from lead while also reducing levels of medical care and the quality of nutrition. The net effect of lower income and less medical care could be a reduction in infants’ prospects.16 Mary E. Becker, From Muller v. Oregon to Fetal Vulnerability Policies, 53 U.Chi.L.Rev. 1219, 1229-31 (1986).
Nothing in the record shows the net risks. When asked at oral argument whether Johnson attempted to determine net effects before adopting its policy, counsel said no. He continued, in effect: “If there is a greater risk, that’s not our concern.” A “not on our watch” position is the bureaucrat’s shelter and not becoming if, as Johnson so earnestly maintains, it is concerned about the welfare of unborn children. These helpless beings do not know or care about the source of risk; they would care only about its aggregate level. Surely Title VII does not allow an employer to adopt a policy that simultaneously makes both women and their children worse off.
Mediation through a Single Sex. Is the risk to the child transmitted by one sex only? If “the risk” is defined as risk caused by lead entering the fetus’s blood via the placenta, it is by definition confined to one sex. But if we ask instead “Does the presence of lead in a parent’s blood pose a risk to the fetus?”, then the evidence conflicts. The broader perspective is the correct one when aggregate levels of risk are the proper concern.
Three affidavits in the record, and papers in medical journals, maintain that lead in the blood creates risks for offspring of both male and female employees. The American Public Health Association and other medical groups have filed a brief as amici curiae marshaling an impressive array of studies linking lead with injury to the male reproductive system, and thence to offspring. E.g., Christopher Winder, Reproductive Effects of Occupational Exposures to Lead: Policy Considerations, 8 NeuroToxology 411 (1987); Herbert L. Needleman & David Bellinger, Commentary, Environmental Research (1988). Most of the data come from animal studies, but some human studies suggest that the effects occur in our species too. OSHA concluded that lead in men as well as women is hazardous to the unborn. The medical surveillance guidelines published as an appendix to the agency’s lead control regulations state:
Exposure to lead can have serious effects on reproductive function in both males and females. In male workers exposed to lead there can be a decrease in sexual drive, impotence, decreased ability to produce healthy sperm, and sterility. Malformed sperm (teratosper-mia), decreased number of sperm (hypos-permia), and sperm with decreased motility (asthenospermia) can all occur. Tera-tospermia has been noted at mean blood levels of 53 ¡ug/100 g and hypospermia and asthenospermia at 41 /¿g/100 g. Furthermore, there appears to be a dose-response relationship for teratospermia in lead exposed workers.... [Bjecause of the demonstrated adverse effects of lead on reproductive function in both the male and female as well as the risk of genetic damage of lead on both the ovum and sperm, OSHA recommends a 30 jiig/100 g maximum permissible blood lead level in both males and females who wish to bear children.
*91929 C.F.R. Part 1910, pp. 833-34 (1987). Perhaps OSHA is wrong; its findings do not bind Johnson in this litigation. But its view has been sustained once after rigorous attention. United Steelworkers of America v. Marshall, 647 F.2d 1189, 1256-58 (D.C.Cir.1980) (describing as “abundant” the evidence supporting the belief that lead injures the reproductive systems of both sexes and sustaining as rational OSHA’s decision that the evidence does not justify excluding women from the workplace). Affidavits and professional articles describing the latest evidence also support OSHA’s judgment. The district court could not properly reject it without holding a trial.
Least of all could a court reject it, as my colleagues do, on the ground that as a matter of law animal studies are not “solid scientific data”, majority op. 889. The District of Columbia Circuit concluded that “virtually undisputed evidence [shows] that at 30 ftg/100 g and above, men suffer ... interference with their ability to produce normal sperm”, 647 F.2d at 1249. The medical profession, like the Food and Drug Administration, will be stunned to discover that animal studies are too “speculative”, majority op. 889, to be the basis of conclusions about risks. Often animal studies are the best foundation for decision. The Supreme Court has concluded that they may be used, e.g., Industrial Union Department v. American Petroleum Institute, 448 U.S. 607, 657 n. 64, 100 S.Ct. 2844, 2871 n. 64, 65 L.Ed.2d 1010 (1980) (plurality opinion). OSHA relied on these very animal studies when establishing its lead rules, as the D.C.Circuit held it may. 647 F.2d at 1257 & n. 97. See also, e.g., Simpson v. Young, 854 F.2d 1429 (D.C.Cir.1988); Public Citizen Health Research Group v. Tyson, 796 F.2d l479, 1489-90, 1492 (D.C.Cir.1986), sustaining risk assessments based on animal studies. Indeed, on occasion animal studies compel responsible agencies to act. Public Citizen v. Young, 831 F.2d 1108 (D.C.Cir.1987).
Less Restrictive Alternatives. An employer cannot close employment opportunities to women in order to protect the next generation if some more modest alternative would do (nearly) as well at protecting the unborn. Johnson’s policy has a striking sweep: no fertile woman can be hired for a job in which any employee has had a blood lead level exceeding 30 /zg/dl anytime during the last year, or in any job that might lead to a promotion to such a job. As a practical matter, this means every industrial job at Johnson’s battery plants. The firm advised its hiring offices to tell women that “we have no openings for women capable of bearing children”. To state the policy is to reveal many less stringent options that might be almost as good at protecting the interests of children.
Women over 40 rarely have children. Why are they forbidden to work? (One of the plaintiffs, 50 years old and divorced when the suit was filed, had nonetheless been excluded from jobs covered by the policy.) Covering them reduces the lead risk to zero, but “zero” is not the only acceptable level.
Many workers in jobs in which some employee has a level exceeding 30 jug/dl have levels less than that. Only Vs of Johnson’s industrial employees exceed the 30 /ig/dl figure. The levels of lead in the blood depend not only on lead in the air but also on personal hygiene. Some workers allow particles to remain on their clothes or person, staging points from which they can be swallowed. Why not allow a woman to enter jobs in which significant fractions of workers sustain levels less than 30 ¡xg/dl, to see whether she can too? Why treat all women as unable to follow good industrial hygiene, or as unwilling to use a respirator, just because Vs of men have elevated levels of lead in the blood? Johnson replies that lead levels usually are greatest shortly after entering a new job, before the employee learns how to reduce the level; the women might become pregnant during these initial weeks or months. True enough, but again this answer assumes that the only acceptable level of risk is zero.
Some workers who start at entry-level jobs with low exposure to lead will be promoted to higher-lead jobs; others will not (or will leave before then). Johnson excludes women even from these safer jobs, although they pose no appreciable risk to offspring. Doubtless Johnson believes that unimpeded lines of progression make *920its operation more efficient, since it can invest more in training women in skills that are transferrable to new jobs within the firm. Yet this form of savings does not count under Title VII. Let us suppose that Johnson had light-lifting and heavy-lifting jobs, joined in a line of progression. Would the inability of many or even all women to perform the heavy-lifting job permit Johnson to exclude all women from the light-lifting job? The exclusion might be desirable from the perspective of business efficiency, but Title VII would not allow it. So, too, Title VII forbids a firm to exclude all women from a line of progression on the ground that some will drop out to have children, rendering their training worthless to the firm. Women have less attachment to the work force than men, making it privately rational for firms to invest less in training them for high-skill jobs; Title VII just as clearly forbids firms to do this. What Johnson has done is little different, and at some cost to itself it can change the fetal protection plan to enlarge women’s opportunities at no cost whatever in risk to the unborn.
Ill
The Wright-Hayes standard is the wrong one. Johnson needed to, and did not, establish that sex is a BFOQ for employment at its battery plants. Yet even given the majority’s decision to adopt the Wright-Hayes standard, the plaintiffs are entitled to a trial. Seven judges of this court have analyzed the conflicting medical evidence and reached their own conclusions about its significance, conclusions at variance with those drawn by the American Public Health Association and the Occupational Safety and Health Administration from the same kind of evidence.17 Judges may be astute students of medical findings, but the presence of thoughtful persons on the other side suggests caution — and at all events appellate judges should not be resolving scientific disputes.
This is the most important sex-discrimination case this circuit has ever decided. It is likely the most important sex-discrimination case in any court since 1964, when Congress enacted Title VII. If the majority is right, then by one estimate 20 million industrial jobs could be closed to women, for many substances in addition to lead pose fetal risks. See note 7 above. Whether that would happen is of course a separate question; legal entitlements need not translate to action. But the law would allow employers to consign more women to “women’s work” while reserving better-paying but more hazardous jobs for men. Title VII was designed to eliminate rather than perpetuate such matching of sexes to jobs.
Title VII requires employers to evaluate applicants and employees as individuals rather than as members of a group defined by sex. The statute has its costs; prenatal injuries are among these. Appeals to the “flexibility” with which the Supreme Court has allocated burdens of proof and persuasion get us nowhere. No amount of “flexibility” justifies sex discrimination without a BFOQ, unless by “flexibility” we mean a prerogative to disregard the statute when it requires decisions antithetical to our beliefs. Although my colleagues refer to many constitutional cases, such as Rostker v. Goldberg, 453 U.S. 57, 101 S.Ct. 2646, 69 L.Ed.2d 478 (1981), for the proposition that sex discrimination sometimes is permissible, cases showing that Congress may authorize sex-based decisions hardly shows that in this instance it did. Title VII forbids rather than requires resort to sex as a basis of decision.
Risk to the next generation is incident to all activity, starting with getting out of bed. (Staying in bed all day has its own hazards.) To insist on zero risk, which the *921court says Johnson may do, is to exclude women from the industrial jobs that have been a male preserve. By all means let society bend its energies to improving the prospects of those who come after us. Demanding zero risk produces not progress but paralysis.' Defining tolerable risk, and seeking to reduce that limit, is more useful — but it is a job for Congress or OSHA in conjunction with medical and other sciences. Laudable though its objective be, Johnson may not reach its goal at the expense of women.
. Two other cases often are treated as bearing on the validity of such policies: Hayes v. Shelby Memorial Hospital, 726 F.2d 1543 (11th Cir. 1984); Zuniga v. Kleberg County Hospital, 692 F.2d 986 (5th Cir.1982). Both involved x-ray technicians fired immediately after their employers discovered that they had become pregnant, and not policies that disqualified women generally. I therefore put Hayes and Zuniga to one side for the time being, although they return in Part II.
. The policy statement, Notice No. n915.034 (Oct. 7, 1988), says at p. 1 n. 1: "three courts of appeals have considered cases involving fetal hazards ..., thereby helping in the development of an analytical framework for deciding such cases. This policy statement is an elaboration of that framework_” See also p. 4: “the Commission follows the lead of every court of appeals to have addressed the question".
. Although some 9% of all fertile women become pregnant each year, the birth rate for blue collar women over 30 is about 2%, and of working women 45-49 only 1 in 5,000 becomes pregnant in a given year. The data are collected in Mary E. Becker, From Muller v. Oregon to Fetal Vulnerability Policies, 53 U.Chi.L.Rev. 1219, 1233 (1986). The record does not reveal the birth rate for Johnson’s female laborers, but it must be lower given Johnson’s strenuous efforts to discourage pregnancy among those exposed to lead.
. One of the children has an elevated level of lead in the blood, but this has not produced an identifiable problem. It is hard, however, to link outcomes such as learning disabilities to lead, since learning disabilities could have many other causes, and it is therefore hard to show in an individual case (as opposed to statistically) that lead injured the child.
. Health care expenses for women are substantially higher than those for men even before taking into account the costs of pregnancy. Department of Health and Human Services, A Summary of Expenditures and Sources of Payment for Personal Health Services From the National Medical Care Expenditure Survey, Data Preview 24 at 11 (May 1987). The PDA requires employers to pick up these higher medical costs, plus the medical costs of pregnancy, if they offer health care to male employees. And after Manhart employers that offer pensions must contribute more per month per female employee than per male employee.
. The survey concluded that 93-95% of the adult population is scientifically illiterate. “Asked whether the Earth goes around the Sun or the Sun around the Earth, 21 percent replied incorrectly; 7 percent said they did not know.... On other questions, the survey found that 43 percent said correctly that electrons, which are components of atoms, are smaller than atoms; 20 percent said they were larger and 37 percent said they did not know which was larger.... Of those surveyed, 76 percent answered correctly that light travels faster than sound; 19 percent mistakenly said that sound moves faster." New York Times, Oct. 25, 1988, § C p. 10 col. 4. The state of knowledge may be even worse in the United Kingdom: "Only 34% of Britons know that the earth goes around the sun (rather than vice versa) and that it takes a year to do so. Almost one-fifth are presumably busy nailing their furniture to the floor as they imagine themselves hurtling around the sun once a day. And they are confused about what the universe is: given a list comprising solar system, galaxy, earth, universe, and sun, only 53% thought that the universe was the largest item. Some 7% thought that the earth was. Asked if nuclear-power stations cause acid rain, almost half the respondents said yes, and only 34% knew that they did not.... So much for environmental awareness.... Some people seem unworried by radiation: 13% think that radioactive milk can be made safe by boiling it.” The Economist, July 15, 1989, p. 84 col. 2.
.Fifteen to twenty million jobs is the estimate of the Bureau of National Affairs in Pregnancy and Employment 57 (1987), limited to injury caused by chemicals. Cases such as Hayes and Zuniga show that many additional women are affected by restrictions placed on other jobs, such as the x-ray technician jobs that exposed embryos to radiation. Concern about emissions from computers and their terminals has led to proposals that could restrict access even to traditional office jobs.
. The EPA, which administers the Toxic Substances Control Act, 15 U.S.C. §§ 2601-54, also may have authority over the subject. See Note, Getting Beyond Discrimination: A Regulatory Solution to the Problem of Fetal Hazards in the Workplace, 95 Yale L.J. 577, 592-98 (1986).
. For a review of the evidence see William L. Marcus & C. Richard Cothern, The Characteristics of an Adverse Effect: Using the Example of Developing a Standard for Lead, 16(4) Drug Metabolism Reviews 423 (1985-86). Dr. Michael Silverstein, a board certified specialist in occupational medicine, testified in a deposition that the risks of a given level of lead in the blood are no greater to the fetus and children than to adults. He supported this position with an affidavit to which he attached three scientific studies. Dr. Marvin S. Legator, Director of the Division of Environmental Toxicology at the University of Texas, gave a deposition agreeing with this view, as did Dr. Ellen Silbergeld, a Senior Scientist in the Toxic Program of the Environmental Defense Fund.
.Dr. J. Julian Chisolm, Director of the Lead Program at the John F. Kennedy Institute in Baltimore, one of Johnson’s witnesses, testified in a deposition that lead does not cross the placenta until late in the pregnancy and that the adverse effect of lead in a mother’s blood does *916not begin until the last half of the third trimester. Dr. Silbergeld took a similar view.
. According to National Research Council Committee on Lead in the Human Environment 59-60 (1980), almost all lead stored in the body outside of bones is eliminated within four to six weeks of exposure. Dr. Silbergeld used the higher figure of 100 days for the elimination of lead from blood and soft tissues. Lead is released from bones much more slowly, but lead-207, the most common form they contain, is less active biologically. National Research Council on Biologic Effects of Atmospheric Pollutants 68 (1972). Even persons who have no occupational exposure ingest or inhale lead from other sources (paint, leaded gasoline, etc.), so that the actual blood lead level may drop slowly, but these sources of lead threaten offspring whether or not the mother works in Johnson’s plants. It is possible that lead builds up in the placenta during early pregnancy, while the mother's blood lead level is highest, but the extent of this effect is unknown. The net effects of these offsetting processes are not quantified in this record.
. E.g., Surgeon General, The Health Consequences of Smoking for Women 191-239 (1980) (carbon monoxide, nicotine, and hydrogen cyanide from smoking cross the placenta, leading to miscarriages, reduction in fetal weight, reduced cognitive development, and Sudden Infant Death Syndrome); Alcohol, Drug Abuse, and Mental Health Administration, Fetal Alco-hoi Syndrome and Other Effects of Alcohol on Pregnancy Outcome (Sixth Special Report to the U.S. Congress on Alcohol and Health) (1987) (maternal drinking is associated with miscarriage, mental retardation of the child, low birth weight, and a characteristic set of birth defects known as "fetal alcohol syndrome”; collecting sources); Ernest L. Abel, Fetal Alcohol Syndrome and Fetal Alcohol Effects (1984) (collecting data on the effect of maternal drinking on the fetus); Richard B. Everson et al., Quantitative Associations Between DNA Damage in Human Placenta and Maternal Smoking and Birth Weight, 80 J.Nat.Cancer Inst. 567 (1988) (maternal smoking damages DNA of fetus); Steven H. Fox et al., Perceptions of Risks of Smoking and Heavy Drinking during Pregnancy: 1985 NHIS Findings, 102 Public Health Reports 73 (1987). Miscarriages, low birth weight, and retarded cognitive development, common consequences of maternal smoking and drinking, resemble the consequences of maternal lead but appear to be more common.
. Centers for Disease Control, Smoking-Attributable Mortality and Years of Potential Life Lost — United States, 1984, 36 Morbidity and Mortality Weekly Report 694 (Oct. 30, 1987).
. See Edmund A.C. Crouch & Richard Wilson, Risk/Benefit Analysis 173-83 (1982); Bruce N. Ames, Renae Magaw & Lois Swirsky Gold, Ranking Possible Carcinogenic Hazards, 236 Science 271 (1987); John Tierney, Not To Worry, *917Hippocrates 29 (Jan/Feb 1988); John F. Mor-rall III, A Review of the Record, Regulation 25 (Nov/Dec 1986).
. There is a tradition in public health of resolving doubts by assuming that risks exist until they can be disproved. As the uncertainties are substantial, this process often produces measures of risk that appear to be substantial. Yet whether to assume that the maximum likely hazard will come to pass — a process known as “conservative risk assessment" — is itself a political question. A court would be obliged to try to produce the most accurate, rather than the most conservative, assessment, for resolving doubts in one direction only produces inaccurate comparisons of poorly-understood risks (which will be overstated) against well-understood risks (which would be accurately stated). See Albert L. Nichols & Richard J. Zeckhauser, The Perils of Prudence: How Conservative Risk Assessments Distort Regulation, Regulation 13 (Nov/Dec 1986). Yet in an effort to measure risk accurately a court would get little aid from existing studies, often tailored to fulfilling regulatory demands for a "conservative” bias.
. The majority says that “[t]his issue bears no relevance to Johnson Controls’ employment practices", majority op. 889 n. 28, because Johnson protected the salary and benefits of women transferred out of jobs in 1982. But Johnson does not offer women excluded from these jobs in 1983 or later the salary and benefits they could have earned in them, and it also does not protect the income and benefits of employees who because of the policy cannot exercise their seniority (bumping) rights to avoid layoffs.
. The amici curiae whose view of the evidence differs from that of the majority include the American Public Health Association, the American Society of Law and Medicine, the Planned Parenthood Federation of America, several toxicologists and physicians who have published scholarly papers on the subject (some of which Johnson relied on, to the consternation of the authors who appear before us to denounce its interpretation of their research), and Eula Bing-ham, a professor of environmental health who was also the Assistant Secretary of Labor for OSHA at the time that agency published its lead regulations. Whether we think these persons and groups right or wrong, we should not suspect them of insufficient devotion to the health and welfare of the next generation.