To evaluate the variability in the reported accuracy of the exercise electrocardiogram (ECG) for predicting severe coronary disease, meta analysis was applied to 60 consecutively published reports comparing exercise-induced ST depression with coronary angiographic findings. The 60 reports included 62 distinct study groups comprising 12,030 patients who underwent both tests. Both technical and methodologic factors were analyzed. Wide variability in sensitivity and specificity was found (mean sensitivity 81% [range 40% to 100%, SD 12%]; mean specificity 66% [range 17% to 100%, SD 16%]). All three variables found to be significantly and independently related to sensitivity were methodologic (the exclusion of patients with right bundle branch block, the comparison with another exercise test thought to be superior in accuracy and the exclusion of patients taking digitalis). Exclusion of patients with right bundle branch block and comparison with a "better" exercise test were both significantly associated with sensitivity for the prediction of triple vessel or left main coronary artery disease. Adjustment of exercise-induced ECG changes for changes in heart rate was strongly associated with the specificity for critical disease (partial R2 = 0.436, p = 0.0001).