We compared the results of coronary artery bypass surgery with those of medical therapy alone in 1491 nonrandomized patients 65 years of age or older. Cumulative survival at six years (adjusted for major differences in important base-line characteristics) was 79 per cent in the surgical group and 64 per cent in the medical group (P less than 0.0001). At five years, chest pain was absent in 62 per cent of the surgical group and 29 per cent of the medical group (P less than 0.0001). Analysis by the Cox proportional-hazards model suggested an independent beneficial effect of surgery on survival (P less than 0.0001). Patients were divided into risk quartiles on the basis of preoperative predictors of survival identified by the Cox model. Surgical benefit was greatest in "high-risk" patients (those in the two quartiles containing patients with the poorest prognosis). Among 234 "low-risk" patients with mild angina, relatively good ventricular function, and no left main coronary artery disease, there was no survival difference between those treated medically and those treated surgically. We conclude that in specific higher-risk subsets of non-randomized patients 65 years of age or older, coronary bypass surgery appeared to improve survival and symptoms in comparison with medical therapy alone. These conclusions must be tempered by consideration of the limitations of nonrandomized studies, particularly since patients in the two treatment groups differed substantially with regard to important base-line characteristics.