Multivariate analysis has been proposed to enhance diagnostic accuracy of the exercise test in coronary artery disease. To quantify the improvement given by multivariate analysis in comparison with ST segment depression alone during exercise test, 558 men without previous myocardial infarction were studied retrospectively. All the patients underwent a symptom-limited Bruce protocol with computer-averaged recordings in V5, aVF, V2. Coronary angiography was performed within the following 90 days. Prevalence of coronary artery disease (diameter narrowing of 50% or greater) was 0.56. Among 12 clinical and exercise parameters studied by stepwise multivariate analysis, five were found to reach the maximal accuracy: (1) exercise duration, (2) history of typical angina, (3) typical angina during the test, (4) age, (5) maximal heart rate. In comparison with ST depression, multivariate analysis significantly improves accuracy (74.6 vs. 66.8%, P less than 0.01) by increasing sensitivity (0.68 vs. 0.59, P less than 0.05) without affecting specificity (0.83 vs. 0.76, NS). Thus, in a group of men without infarction and a low prevalence of coronary artery disease, multivariate analysis with five easily collected variables is more accurate than ST segment depression alone during exercise. In addition, ST depression did not appear as discriminant as exercise duration for diagnostic purposes. This finding emphasises the importance of performing a symptom-limited exercise test.