Attention is drawn to the possible distortion of results in randomized trials of exercise‐centered rehabilitation following myocardial infarction, due to a poor compliance of patients with the assigned form of treatment. The possibility of a post‐hoc classification of exercise behavior is explored, with particular reference to data obtained in the Ontario Exercise‐Heart Collaborative Study, which compared the therapeutic efficacy of high‐intensity exercise (HIE) and light‐intensity exercise (LIE) programs. On the basis of the heart rate observed during cycle ergometer exercise at an oxygen consumption of 1.25 l‐min‐1, subjects were assigned to a HIE‐trained subgroup (92 of the patients originally assigned to HIE, with a decrease in exercise heart rate>5 beats‐min‐1during the first year of observation), and a LIE‐control subgroup (102 of the patients originally assigned to LIE, with a decrease in exercise heart rate<5 beats‐min‐1over the first year). After this reclassification, the odds ratio relating the risk of reinfarction for HIE‐trained to LIE‐control was decreased, but still did not differ significantly from 1.00 (mean, 0.74; confidence interval, 0.20 to 2.65). Moreover, reasons are advanced to suggest that because of a selection of subjects, reclassification may have reduced the risk of recurrence for the HIE group, while increasing the risk for the LIE group. It is concluded that practical obstacles will preclude formal statistical proof of the exercise hypothesis for the postcoronary patient for the forseeable future.