In 1988, this court decided Aman v. Federal Aviation Administration, 856 F.2d 946 (7th Cir.1988) (Aman I), in which current and former airline captains sought review under 49 U.S.C.App. § 1486(a) of a Federal Aviation Administration (FAA) order denying a petition for exemptions from FAA regulation § 121.383(c) (the “age sixty rule”). The age sixty rule prohibits flights under Part 121 of FAA regulations, including commercial flights of aircraft seating more than thirty passengers, from taking off under the command of pilots age sixty or older. See 14 C.F.R. §§ 121.1, 121.383(c). While the FAA is empowered to grant exemptions to this rule if it “finds that such action would be in the public interest,” see 49 U.S.C.App. § 1421(e) (the public interest standard), no exemptions have ever been granted.
After recounting the age sixty rule’s tortured thirty year history and establishing the appropriate “substantial evidence” standard of review, this court in Aman I carefully demarcated petitioners’ two claims. The first was that pilots, age sixty or older, meeting petitioners’ proposed battery of physical and psychological tests (the protocol), were no more likely to cause accidents due to sudden incapacitation or undetected deterioration of piloting skills than other pilots. The FAA rejected this claim, finding that it was not in the public interest to grant exemptions when petitioners’ protocol did not surely reduce all incremental risks associated with the aging process. Although this court noted that it might well have concluded otherwise as a matter of first impression, it held that substantial evidence supported the FAA’s finding.
Petitioners’ second claim was that the flying experience gained by allowing pilots age sixty and older to fly offset any increased risk of accident due to sudden incapacitation or skill deterioration, and that granting limited exemptions effectively produced a net increase or, at least, no net decline in safety. The FAA summarily rejected this claim. Unsatisfied with the agency’s cursory treatment, this court vacated the FAA order and remanded the matter to the agency for further proceedings.
On remand, the FAA again refused to grant exemptions, and an order to that effect is presently before us for review. After considering the FAA’s new order and both parties’ somewhat flawed evidence, we cannot justify a conclusion that, on average, experience sufficiently offsets possible age-related impairment of health or skills to clearly guarantee a net constancy or increase in safety. Accordingly, we affirm.
I.
While substantial evidence must support the FAA’s decision, see Aman I, 856 F.2d at 951-52, petitioners have the burden of showing that circumstances justify exemptions from the age sixty rule, especially given the FAA’s discretionary authority to act in this area. Starr v. FAA, 589 F.2d 307, 311 (7th Cir.1978). It is a heavy burden here involving obviously daunting problems of public safety. Age discrimination may form a dimension of the issue, but safety is the dominant and controlling consideration. The fact that it is apparently very difficult to demonstrate any clear conclusion with respect to the trade-off between experience and possible age-related impairment makes the task extremely onerous for the bearer of the burden of persuasion.
Petitioners have presented anecdotal evidence of superannuated pilots performing heroic deeds. Consider, for example, Captain David Cronin, who at age 59, on his second to last scheduled flight, heroically landed a Boeing 747 en route from Honolulu, Hawaii to Auckland, New Zealand after a forward cargo door blew open 17 minutes after take off, opening a huge hole in the *320side of the plane. After two of the plane’s four engines became disabled, Captain Cronin consulted emergency operating procedures which directed him to dive, reduce speed and drop the landing gear. However, 38 years of experience told him that, if that course were followed, the plane would lose too much altitude given its weight and multiple emergency situation. Captain Cronin instead operated many of the controls manually, constantly readjusting his speed and altitude calculations. With the exception of the nine passengers killed when the cargo door blew off, Cronin saved the lives of all passengers and crew aboard, safely landing the disabled plane at a much higher than normal speed. Pet. Ex. 105, 105A, 106, J.A. 278-80. In an appropriate context, we might give considerably more weight to the “anecdotal” evidence of pilots in their late fifties immediately before retirement performing amazing feats of airmanship than presumably would the FAA. As noted at oral argument, were the passengers of Flight 811 asked whether their Captain Cronin should be permitted to continue flying beyond the mandatory retirement age of sixty, few could doubt their answer. Nor are we in a position to say they would be incorrect. In the case before us, however, it is apparently not pilots who have performed aeronautical miracles who have sought exemptions, and we need not consider the arguable entitlement of such “special” pilots to exemptions from the age sixty rule.
While petitioners have thus made some suggestive anecdotal showings and presented impressive expert opinion evidence, they have been unable to develop a persuasive statistical record comparing average risks for pilots in various relevant age categories. Petitioners, relying on figures from the National Transportation Safety Board, presented evidence that pilots age sixty and older had a lower accident rate per 1,000 pilots than pilots in other age groups. Pet. Ex. 71, 72, J.A. 215-16. This evidence, however, failed to account for exposure to risk in terms of hours of flight time. Thus, a pilot who had flown only a relatively few hours in a year and therefore incurred only a reduced risk of accident would carry the same weight as a pilot who flew many hundreds of hours with their greater attendant risks. R. 5; R. 103. Such a study is, of course, of questionable value. In addition, no analysis indicated whether the difference in accident experience by age group was statistically significant, a sort of failure specifically criticized in Aman I. 856 F.2d at 955. Petitioners also presented evidence that allowing pilots, sixty or older, to fly would increase crew experience on the average, but failed to show with any rigor that there was a significant lack of pilot experience in need of correction.
This is not to say that the FAA’s evidence was any more persuasive. The agency relied heavily on an accident experience report by age. category referred to as the Flight Time Study. Like petitioners’ studies this report has serious flaws.1 Perhaps the Flight Time Study’s greatest failing is that the data for pilots under age sixty include millions of relatively safe air carrier miles flown, miles which because of the age sixty rule were unavailable to pilots over sixty.2 In calculating the accident rate for pilots sixty and older, the Flight Time Study divides the number of general aviation accidents by general aviation flight time, the only category open to this group. But for pilots under age sixty, the study divides the number of general aviation accidents by general aviation flight time and, in addition, air carrier operations flight time. Because miles flown in air carrier operations are nearly accident free, *321and millions of these extra miles are included in the figures for younger pilots but not for older ones, the accident rate for all pilots under age sixty is significantly understated compared to the rate for older pilots, whose accident rate is overstated. Indeed, looking at the Flight Time Study’s chart of accident risk for Class I (airline transport) and Class II (commercial) pilots with greater than 5,000 hours total flight time (Pet.App. 42, Fig. 7), the jump in accidents at age sixty to sixty-nine from age fifty to fifty-nine simply looks too large to be credible.3
Even without correcting the Flight Time Study for this disparity in types of current flight hours, the FAA’s own study on its face may in some aspects be construed to support the petitioners’ claims, the raw data supporting a number of different possible conclusions. Consider, for example, the data showing accident rates as a function of both total and recent flight time for Class III pilots (general aviation and student). These data indicate that pilots age 60-69 (even 70 and over) with more than 1,000 hours total flight time and more than 50 hours recent flight time apparently have two of the lowest accident rates of any age groups of pilots in Class III having various indicated combinations of total and recent flight time. These comparisons apply, of course, even with respect to younger pilots in their thirties and forties, whose safety qualifications are generally unquestioned. Flight Time Study, J.A. 551, Fig. 9, J.A. 575-79, 809. More than 1000 hours total flight time and more than 50 hours recent flight time might be a telling statistic if the FAA could, for example, condition exemptions on total and recent flight time. Safety would be advanced, presumably, if the FAA required pilots to have at least a total of 1,000 flight hours and at least 50 recent flight hours as a condition for exemption.4
Another arguable flaw in the Flight Time Study is that all pilots in a ten-year age cohort are combined into a single statistic. Thus, a single point represents pilots aged sixty to sixty-nine. Presumably, more exemptions from the age sixty rule would likely be granted to pilots under, say, sixty-five than pilots over that age. Therefore, the cumulation of accidents caused by pilots in their late sixties with accidents caused by pilots in their early sixties may as a practical matter tend to skew the Study.
The FAA also presented evidence of automobile traffic accidents and fatalities related to age. Res.Addendum B. The connection between automobile drivers and pilots itself seems tenuous given the pilots’ training, demonstrated proficiency, medical fitness, etc. Some of the FAA’s evidence does not reflect “exposure” and some of it attempts to relate the nonparallel categories of automobile fatalities and aircraft accidents. Res.Br. at 19, n. 17. Because elderly people seem more likely to die as a result of traffic accidents than younger people, the probative value of this latter showing is diminished.
Along with a directive for a more complete consideration of petitioners’ second claim on remand, the FAA was also requested to explain how it could rationally grant exemptions to younger pilots who had suffered from alcohol abuse, heart conditions and the like but not grant exemptions to apparently healthy and proficient pilots over age sixty. The agency’s justification was that, where particular and identifiable health problems are shown, specific medical tests may be conducted to indicate whether the pilot in question can continue to perform. On the other hand, “[ajssessing the risks associated with determining which pilots may fly beyond age 60 concerns detrimental conditions which are unknown.” Res.Br. at 32. Exactly how this *322distinction applies as a practical matter is not entirely clear to us, but neither have the petitioners been able to demonstrate its invalidity either theoretically or practically. Nor have petitioners apparently yet been able to present to the FAA a totally reliable test or group of tests which would reveal with certainty any general deterioration of piloting skills associated with advancing age. See Aman I, supra, at 954. For present purposes, we will not require more of the FAA with respect to the consistency of restoring stricken younger pilots to duty while barring oldsters whose records are impeccable.
II.
Admittedly, petitioners in this case face a Catch-22: from one perspective they cannot get exemptions until they show they can fly large passenger aircraft safely, and they cannot show they can fly such planes safely until they get exemptions. Thus, a valid statistical demonstration of comparative safety records by age seems difficult to obtain unless all age groups are engaged in the same kinds of flying. Since the age sixty and over group may not pilot large passenger transport aircraft, statistical comparisons are suspect. Nevertheless, it was the petitioners’ burden to present persuasive evidence that granting exemptions would not impair safety. While we have seen no compelling evidence that granting exemptions would increase the risk of accident, neither have we seen strong evidence that the experience of the 60-and-over pilot clearly overbears the danger of deterioration of piloting skills (or of sudden incapacitation) associated with the aging process. Where crucial issues of public safety are at stake, we would look for such a showing. Were the FAA to grant exemptions, it (and we) would no doubt be better able to resolve the question before us, but, absent the requisite compelling evidence, we must defer in these circumstances to the expert agency.5
We believe the agency’s order is supported by substantial, albeit certainly not compelling, evidence.6 We reach this conclusion because of the obvious difficulty in attempting to balance on a statistical basis experience against reliable indicators of good health and ability to perform as age advances. Certainly the record abounds with testimonials by experts in both flying and medicine to the experience and judgment of the older aviator and the feasibility of assuring the good health and performance of this kind of pilot through frequent and sophisticated testing. We are certainly not in a position to say that the numerous supporters of the petitioners’ case are wrong. And it is obvious that the FAA must continue and must enhance its efforts to accommodate their points of view. At this time, however, we are not prepared to overrule the agency in a matter of such immense sensitivity as this one.7 The FAA should not take this as a signal that the age sixty rule is sacrosanct and untouchable. Obviously, there is a great body of opinion that the time has come to move on. *323The agency must give serious attention to this opinion.
Accordingly, the order of the Federal Aviation Administration is
Affirmed.
. Petitioners point out that the Flight Time Study is not even relevant since that study, as originally presented seven years ago, was based on data from 1976 to 1980 and was never intended to justify the age sixty rule. Numerous comments of record from various experts, even some from the FAA, state that the study should not be relied on as determinative — or even probative — on the question of the continued validity of the age sixty rule. As discussed infra, even if the study is relevant, it is only of very limited usefulness.
. This vast discrepancy in flight exposure for the group studied is one reason the petitioners presented figures on a per-pilot basis without reference to hours flown.
. The chart which shows petitioners’ attempt to recalculate these numbers and correct for the safe miles unavailable to pilots over age sixty appears to err in exactly the opposite direction. Pet.Br. at 17. Again, neither party performed statistical analysis which would indicate whether the differences were significant.
. While Class III is not the universe from which air carrier pilots would be selected, it is the category where the ratio of accidents to flight time is subject to the least distortion.
. In Aman I, we also suggested that there might be administrative or economic reasons for the agency’s policy of withholding exemptions from the age sixty rule and invited the FAA to discuss these aspects of the problem if they were relevant. The agency has elected on remand not to address non-safety concerns. We think that administrative and economic factors could hardly be entirely absent. But, since the petitioners have been unable to carry the heavy burden of establishing a case for exemptions from the rule on a safety basis alone, we need not speculate about the economic and administrative aspects.
. We would reiterate that we are not, as the dissent argues, holding that "every airline pilot, on his or her 60th birthday, and regardless of physical condition or experience, becomes a significantly greater safety hazard than before____’’ Infra at 326 (emphasis in original). The grant of specific exemptions from the age 60 rule, and not the validity of the rule itself, is at issue here. In any event, line-drawing may inevitably involve some arbitrariness.
.The principal prescription of the dissent would be merely to remand "for consideration of the adoption of regulations establishing ascertainable and meaningful standards to govern the granting of at least some exemptions to the age 60 rule." Infra at 326-327. Of course, as our opinion suggests, we too would look with favor upon this sort of consideration by the FAA. However, this consideration, because of the inevitable delays, would likely be of slight help to these petitioners.