This multicenter study examines in detail the spontaneous increase in pump flow at fixed speed that occurs in exercise. Eight patients implanted with the VentrAssist rotary blood pump were subjected to maximal and submaximal cycle ergometry studies, the latter being completed with patients supine and monitored with right heart catheter and echocardiography. Maximal exercise studies conducted in each patient at three different pump speeds on separate days established initially the magnitude and consistency of increases in pump flow that correlated well with changes in heart rate. However, there was considerable variation, coefficients of variation for mean heart rate and pump flow being 47.9 and 49.3%, respectively. Secondly, these studies indicated that increasing pump flows caused significant improvements in maximal exercise capacity. An increase of 2.1 L/min (35%) in maximum blood flow caused 12 W (16%) further increase in achievable work, 1.26 (9.3%) mL/kg/min in maximal oxygen uptake, and 2.3 (23%) mL/kg/min in anaerobic threshold. Mean increases in lactate were 0.85 mm (24%), but mean B-type natiuretic peptide fell by 126 mm, (-78%). From submaximal supine exercise studies, multiple linear regression of pump flow on factors thought to underlie the spontaneous increase in pump flow indicated that it was associated with increases in heart rate (P = 0.039), pressure gradient across the left ventricle (P = 0.032), and right atrial pressure (P = 0.003). These changes have implications for the recently reported Starling-like controller for pump flow based on pump pulsatility values, which emulates the Starling curve relating pump output to left ventricular preload. Unmodified, the controller would not permit the full benefits of this effect to be afforded to patients implanted with rotary blood pumps. A modification to the pump control algorithm is proposed to eliminate this problem.