Simple conceptual ideas about cardiac resynchronization therapy assume that biventricular (BiV) pacing results in collision of right and left ventricular (LV) pacing-derived wavefronts. However, this concept is contradicted by the minor reduction in QRS duration usually observed. We investigated the electric mechanisms of cardiac resynchronization therapy by performing detailed electric mapping during extensive pacing protocols in dyssynchronous canine hearts. Studies were performed in anesthetized dogs with acute left bundle-branch block (LBBB, n=10) and chronic LBBB with tachypacing-induced heart failure (LBBB+HF, n=6). Activation times (AT) were measured using LV endocardial contact and noncontact mapping and epicardial contact mapping. BiV pacing reduced QRS duration by 21±10% in LBBB but only by 5±12% in LBBB+HF hearts. Transseptal impulse conduction was significantly slower in LBBB+HF than in LBBB hearts (67±9 versus 44±16 ms, respectively), and in both groups significantly slower than transmural LV conduction (≈30 ms). In both groups QRS duration and vector and the epicardial AT vector amplitude and angle were significantly different between LV and BiV pacing, whereas the endocardial AT vector was similar. During variation of atrioventricular delay while LV pacing, and ventriculo-ventricular delay while BiV pacing, the optimal hemodynamic effect was achieved when epicardial AT and QRS vectors were minimal and endocardial AT vector indicated LV preexcitation. Due to slow transseptal conduction, the LV electric activation sequence is similar in LV and BiV pacing, especially in failing hearts. Optimal hemodynamic cardiac resynchronization therapy response coincides with minimal epicardial asynchrony and QRS vector and LV preexcitation.