This large-scale study examined the ability of stepwise discriminant analysis of clinical, exercise, and thallium tomographic data to detect high-risk patients with three-vessel or left main disease. There were 834 patients, 229 with three-vessel or left main disease (group 1) and 605 (group 2) with either two-vessel disease (n = 236), one-vessel disease (n = 195), or no coronary artery disease (n = 174). The two groups were different in age, exercise heart rate, ST segment depression during exercise, exercise systolic blood pressure, abnormal thallium scans, reversible perfusion defects, extent of thallium abnormality, number of vascular territories with perfusion abnormalities, left ventricular cavity dilatation, and increased lung thallium uptake. On multivariate stepwise discriminant analysis, only three variables were independent predictors of high risk. These included multivessel thallium abnormality (F = 107, p < 0.001). exercise heart rate (F = 27, p < 0.001), and ST segment depression (F = 8, p < 0.001). Based on these three variables, patients could be stratified into three categories with different prevalences of left main or three-vessel disease; the prevalence was 53% in 239 patients, 24% in 271 patients, and 12% in 324 patients. Thus high-risk patients with left main or three-vessel disease can be identified by exercise thallium tomographic imaging that uses a model based on stepwise discriminant analysis. The thallium data are far more powerful than the clinical or treadmill exercise data. © 1993.