In a group of 207 patients undergoing coronary angiography, myocardial scintigrams have been performed at rest (n = 95), after exercise testing (n = 50) or after dipyridamole vasodilation (n = 62). Pictures recorded in anterior and 45 degrees left anterior oblique projections have been analysed according to a model which divides each projection into 5 territories. Thallium uptake has been visually assessed using a three grade scale: 0 = normal uptake, + = sightly abnormal uptake, ++ = evident hypoactivity. With reference to angiography, sensitivity and specificity for prediction of coronary artery disease have been calculated in a classical way using three different criteria of positivity of increasing severity, and by a computerized method based on discriminant analysis. In the first case, sensitivity and specificity were highly dependent one on the other and, as could be expected, varied in opposite directions, a high sensitivity (89 to 100%) being only achieved by accepting a lack of specificity (33 to 57%). Discriminant analysis, on the contrary, provided for each type of protocol: rest, stress, dipyridamole and redistribution, a single optimized combination of sensitivity (74 to 87%) and specificity (92 to 100%). The corresponding point, when located on the receiver operating characteristics (ROC) diagram, demonstrated a clear improvement in diagnostic accuracy. The same method has been applied for discrimination, not only between normals and abnormals, but also between normals and patients with and without myocardial infarction (correct diagnosis in 50 to 68% of cases), and between normals and patients with single-, double- or triple-vessel disease (correct diagnosis in 17 to 65% of cases) with a high specificity (correct classification of normals in 86 to 100% of cases).