An evaluation was made of the performance of an on-line computer system for the detection of premature ventricular contractions. Following a rigidly defined protocol, 34 patients having an average PVC rate of almost one out of every twelve beats were studied, and approximately fifty thousand beats were analyzed independently by human observer and correlated with computer analysis. In the population studied, 78% of the PVCs were detected accurately. In addition, no patient was encountered having a PVC which was always missed by the computer. No false positive PVCs occurred in half the cases, and in 31 cases, only 39 computer labeled "PVCs" could be clearly called false positive PVCs based on information in the single-lead electrocardiogram. The patients studied had a variety of different rhythm complications including atrial fibrillation and premature ventricular contractions from multiple foci. The factors affecting performance were analyzed and directions for future improvement discussed.