The ability to noninvasively detect coronary artery disease (CAD) in patients undergoing diagnostic cardiac catheterization was studied using a signal-averaged electrocardiogram. An initial study of 13 patients revealed that a QRS duration greater than or equal to 100 msec, a root mean square voltage in the terminal 40 msec of the QRS less than 50 microV, and a low amplitude signal (LAS) duration greater than 28 msec were suggestive of CAD. These parameters were then used prospectively to examine 40 consecutive patients with chest pain of undetermined etiology referred for cardiac catheterization. Patients with CAD had significantly longer filtered QRS and LAS durations and lower root mean square voltages compared with patients without CAD. The sensitivity, specificity, and positive predictive value of a single parameter ranged from 62% to 76%, 74% to 89%, and 75% to 87%, respectively. Thus the signal-averaged electrocardiogram may be a useful tool in evaluating patients for the presence of CAD.