A consecutive series of 56 patients with chest pain but no evidence of previous myocardial infarction was prospectively studied by radionuclide ventriculography to determine the value of global and regional radionuclide indices in detecting coronary artery disease. The results were correlated with the clinical judgment of chest pain, the results of the exercise electrocardiogram, and the right heart haemodynamic measurements during exercise. As a result of the criteria for entry, the study group was representative of the population seen in such a clinical setting. Only 25% of patients had coronary artery disease. The predictive power of radionuclide ventriculography was limited. The conventionally used criterion that normal subjects have an increase in left ventricular ejection fraction of at least 5% with exercise provided only 78% sensitivity and 57% specificity. Fourier analysis and visual interpretation of radionuclide studies wrongly diagnosed three out of 10 patients with extensive disease requiring surgery. These results suggest that radionuclide ventriculography is of limited value in the non-invasive diagnosis of coronary artery disease.