The radionuclide ventriculographic exercise response was evaluated in three patient populations representing alternative referent standards for cardiac normality: patients with normal coronary arteriograms, healthy volunteers, and uncatheterized patients with a low probability of coronary artery disease. Disease probability was determined by Bayesian analysis of age, sex, symptoms, and the results of cardiac fluoroscopy, exercise electrocardiography, or thallium scintigraphy. A wide range of ventriculographic responses was noted in the 62 catheterized normal patients; 21 (34%) had an abnormal ejection fraction response and 22 (35%) had an abnormal wall motion response. In contrast, the ejection fraction and wall motion responses were normal in the 9 volunteers. In 90 patients (18 catheterized and 72 uncatheterized) who had low disease probability ( < 1%), abnormal responses were rare; the ejection fraction response was abnormal in only 7% and the wall motion response was abnormal in 8%. Thus, these three populations are not equivalent referent standards of normality. Volunteers and patients with low disease probability provide too strict a standard, and their use can overestimate test specificity; catheterized normal patients, on the other hand, provide too lenient a standard, and their use can underestimate test specificity.