OBJECTIVE To compare the longitudinal changes in maximal aerobic capacity (VO2max) in healthy middle aged and older athletes and sedentary men. METHODS A cohort study with mean follow-up of 8.7 years (range 4.0-12.8). METHODS Outpatient research at a tertiary hospital. METHODS Forty-two healthy, middle aged, and older athletes (initial age 64 +/- 1 year) and 47 healthy sedentary men of comparable age recruited for research studies. METHODS VO2max during a maximal treadmill test. RESULTS At baseline, the cross-sectional rates of decline in VO2max with age (slope) were virtually identical in the athletes and sedentary men (-0.42 versus -0.43 mL x kg(-1) x min(-1) x year(-1)). At follow-up, the VO2max had declined by 11.9 +/- 1.1 mL x kg(-1) x min(-1) (22%) in the athletes, a crude average rate of -1.4 +/- 0.14 mL x kg(-1)x min(-1) x year(-1). By comparison, the VO2max declined by 4.4 +/- 0.6 mL x kg(-1) x min(-1) (14%) in the sedentary men, a crude average rate of change of -0.48 +/- 0.07 mL x kg(-1) x min(-1) x year(-1). Therefore, the observed absolute rate of longitudinal decline in VO2max in the athletes was triple that of the sedentary men (P= .001) and significantly greater than the decline predicted by their baseline cross-sectional data (P= .001). Post hoc analyses of the longitudinal data in the athletes based on the training regimens over the follow-up period demonstrated that the seven individuals who continued to train vigorously ("high training") had no significant decline in VO2max (0.28% change in VO2max per year). By contrast, the VO2max declined by 2.6% per year in the "moderate training" group (N=21), 4.6% per year in the "low training" group (N=13), and 4.7% per year in the two individuals who developed cardiovascular disease. CONCLUSIONS The longitudinal decline in VO2max in older male endurance athletes is highly dependent upon the continued magnitude of the training stimulus. The majority of the athletes reduced their training levels over time, resulting in longitudinal reductions in VO2max two to three times as large as those predicted by cross-sectional analyses or those observed longitudinally in their sedentary peers.