The efficiency of left ventricular assist devices (LVAD) depends on the capacity of the inflow cannula to drain blood into the pump. Left atrial (LA) and left ventricular (LV) sites are compared in an animal model mimicking different hemodynamic situations. Three calves (56.3±5.0 kg) monitored with ECG, arterial pressure and central venous pressure (CVP) were equipped with a Thoratec LVAD used in fill-empty mode. A regular CPB pump was used as a RVAD (jugular vein/pulmonary artery). CVP was adjusted to defined preload conditions by storage (or perfusion) of blood into (or from) the venous reservoir. We compared the LA and LV inflow cannulation sites, used separately or simultaneously, on the output of the LVAD, tested alone on a beating heart, and with the CPB (BiVAD) on a beating then fibrillating heart. In every conditions, increase of CVP highlighted the differences of LVAD's performance due to the sites of inflow cannulation. For a CVP of 14 mmHG, LVAD used on a beating heart was better with the LV (4.5±0.4 l/min) as compared to the LA cannulation (3.9±0.4 l/min, p < 0.01). Results were similar with a BiVAD used on a beating heart (4.9±0.2 1/min for LV vs. 3.8±0.6 l/min for LA cannulation, p < 0.001). However, with a fibrillating heart, LA cannulation allowed better LVAD performance (2.9±0.3 vs. 2.4±0.1 l/min, p < 0.001). Using both LV and LA inflow cannulae simultaneously, did not improve the LVAD output in none of the situations. LV cannulation increases LVAD performance in beating hearts. However, because the fibrillating heart represents the extreme of a failing heart, the use of a LA site for inflow cannulation must be considered for a better BiVAD performance.