The optimal bandpass filter for signal averaging of the surface QRS complex to detect late potentials is undefined. A study was conducted in 87 patients; 25 (mean age 34 ± 10 years) were normal (group I), 29 (60 ± 20 years) had organic heart disease without ventricular tachycardia (group II) and 33 (62 ± 15 years) had sustained ventricular tachycardia (group III). In all patients signal-averaged electrocardiography (200 beats) was performed using a sharp, bidirectional filter and data analyzed using the following 7 high-pass filter settings: 10, 15, 20, 25, 40, 80 and 100 Hz. For each filter the duration of the signal-averaged QRS complex, the low-amplitude signals of less than 40 μV and the root-mean-square voltage of the terminal 40 ms (RMS-40) were determined. Normal values for each filter were determined from group I patients. In all 3 groups, quantitative signal-averaged variables were filter dependent. There was a progressive and marked decrease in RMS-40 and a progressive and marked increase in low-amplitude signal duration as the high-pass filtering was increased from 10 to 100 Hz. In contrast, high-pass, filter-dependent changes in signal-averaged QRS duration were less marked. The sensitivity and specificity for each filter using RMS-40 as the index of late potentials in separating group III patients from group II patients were: 10 Hz—64% and 52%; 15 Hz—57% and 72%; 20 Hz—57% and 76%; 25 Hz—42% and 90%; 40 Hz—61% and 83%; 80 Hz—88% and 69%; and 100 Hz—79% and 62%, respectively. When 2 or more variables were taken into account, sensitivity did not increase relative to a single variable; however, the specificity increased substantially. In conclusion, the sensitivity and specificity of late potentials in separating patients with ventricular tachycardia from those without is high-pass filter dependent. A level of 25 Hz provides a low sensitivity but the best specificity, whereas 80 Hz provides the best sensitivity but a low specificity. A filter of 40 Hz provides sensitivity and specificity intermediate between those at 25 and 80 Hz. Thus 25, 40 and 80 Hz are optimal bandpass filter levels for signal averaging of the QRS complex to detect late potentials and each offers advantages of sensitivity and specificity over the other.