This study examined the predictors of survival of patients with severe left ventricular dysfunction (resting ejection fraction ≤35%). There were 132 patients aged 57 ± 13 years (mean ± SD) who underwent symptom-limited upright exercise testing for the evaluation of mild stable symptoms of angina pectoris or dyspnea or both (New York Heart Association Classification class II). Radionuclide angiograms were obtained at rest and peak exercise. During a follow-up of up to 42 (mean 24) months 23 patients died of cardiad causes (17%). Univariate survival analysis using the Cox model identified several predictors; multivariate analysis, however, showed that (1) among the nonexercise variables, the left ventricular enddiastolic volume index was the best predictor (χ2 = 10, p = 0.002); (2) among the nonscintigraphic exercise variables, the exercise duration was the best predictor (χ2 = 7.7, p = 0.006);(3) among the scintigraphic exercise variables, the exercise ejection fraction was the best predictor (χ2 = 16.5, p = 0.0001), and (4) when all important variables were analyzed simultaniously, the exercise ejection fraction ramained the best predictor (χ2 = 14.8, p = 0.0001). The survival in patients with exercise ejection fraction >30% was 100% at 12 months and 96% at 42 months while the survival of patients with exercise ejection fraction ≤30% was 96% at 12 months and 64% at 42 months(Mantel-Cox, p = 0.0009). Thus, patients with left ventricular dysfuction constitute a heterogeneous group in regard to cardiac performance at rest and during exercise. More importantly, high- and low-risk groups can be identified; the exercise left ventricular ejection fraction is the most important predictor of mortality.