A multivariate risk function was developed on data from all 508 medical patients in the Veterans Administration (VA) randomized study of coronary bypass surgery. The variables, in order of importance, were ST-segment depression on resting ECG, history of myocardial infarction, history of hypertension and New York Heart Association functional classification III or IV. These noninvasive variables have been reported to be risk factors in natural-history studies of coronary heart disease (CHD). Applying the risk function to medical and surgical patients of the 1972-1974 cohort yielded a 5-year probability of dying for each patient. Investigation of treatment effects in approximate terciles obtained by collapsing the probability distribution into low-, middle- and high-risk groups showed that surgery was beneficial for patients in the high-risk tercile even after removal of patients with left main coronary artery disease (17% surgical vs 34% medical mortality at 5 years; p less than 0.01). This finding was accentuated when patients in the 10 hospitals with the lowest operative mortality (3.3%) were compared. Mortality results in the low-risk tercile favored medical treatment (medical vs surgical mortality 7% vs 17%; p less than 0.05). The risk function predicted mortality well not only for te VA medical group, but also for an independent symptomatic CHD population from the University of Alabama arteriography registry. This report further delineates the advantages and limitations of coronary bypass surgery in CHD patients with chronic stable angina.