To examine prospectively the association between regular exercise and the subsequent development of non-insulin-dependent diabetes mellitus (NIDDM). Prospective cohort study including 5 years of follow-up. 21,271 US male physicians participating in the Physicians' Health Study, aged 40 to 84 years and free of diagnosed diabetes mellitus, myocardial infarction, cerebrovascular disease, and cancer at baseline. Morbidity follow-up was 99.7% complete. Incidence of NIDDM. At baseline, information was obtained about frequency of vigorous exercise and other risk indicators. During 105,141 person-years of follow-up, 285 new cases of NIDDM were reported. The age-adjusted incidence of NIDDM ranged from 369 cases per 100,000 person-years in men who engaged in vigorous exercise less than once weekly to 214 cases per 100,000 person-years in those exercising at least five times per week (P, trend, less than .001). Men who exercised at least once per week had an age-adjusted relative risk (RR) of NIDDM of 0.64 (95% Cl, 0.51 to 0.82; P = .0003) compared with those who exercised less frequently. The age-adjusted RR of NIDDM decreased with increasing frequency of exercise: 0.77 for once weekly, 0.62 for two to four times per week, and 0.58 for five or more times per week (P, trend, .0002). A significant reduction in risk of NIDDM persisted after adjustment for both age and body-mass index: RR, 0.71 (95% Cl, 0.56 to 0.91; P = .006) for at least once per week compared with less than once weekly, and P, trend, .009, for increasing frequency of exercise. Further control for smoking, hypertension, and other coronary risk factors did not materially alter these associations. The inverse relation of exercise to risk of NIDDM was particularly pronounced among overweight men. Exercise appears to reduce the development of NIDDM even after adjusting for body-mass index. Increased physical activity may be a promising approach to the primary prevention of NIDDM.