In this study we assessed whether various responses to exercise testing could be quantified in order to derive the probabilities of presence of coronary disease, and if present, to assess its severity. A treadmill score based on the exercise response was determined in 405 patients who had both treadmill tests and coronary angiograms. The score was derived using discriminant function analysis, by weighting and combining depth and configuration of ST depression (downsloping, horizontal or slowly upsioping), timing onset and duration of ischemia, grading ventricular arrhythmias, heart rate and blood pressure change, coexistence of exercise-induced chest pain and sex. The treadmill score was effective in detecting coronary disease (lesions with an > 50% narrowing), with a predictive accuracy (PA) (probability that a subject manifesting a positive test has disease) of 87%, a true negative rate (TNR) (probability of a subject with a negative test having no disease) of 80%, and sensitivity of 94%. The treadmill score also detected severe disease (triple-vessel, main left and/or > 90% proximal occlusion of the left anterior descending artery), with a PA of 73%, TNR of 79% and sensitivity of 82%. We conclude that the exercise response, expressed numerically as a treadmill score, permits analysis of most of the relevant data from exercise testing, increases test accuracy by 10-15% compared with standard criteria for treadmill test interpretation, and enables the derivation of probability statements for presence and severity of coronary disease. The validity of any prediction on the basis of exercise performance may thus be quantitatively judged.