We have found that when suprathreshold cathodal stimuli were applied to the epicardium of canine ventricle, impulse propagation originated at a "virtual cathode" with dimensions greater than those of the physical cathode. We report the two-dimensional geometry of the virtual cathode as a function of stimulus strength; the results are compared with the predictions of an anisotropic, bidomain model of cardiac conduction recently developed in our laboratories. Data were collected in six pentobarbital-anesthetized dogs by using a small plaque electrode sewn to the left ventricular epicardium. Arrival times at closely spaced bipolar electrodes oriented radially around a central cathode were obtained as a function of stimulus strength and fiber orientation. The dimensions of the virtual cathode were determined by linear back-extrapolation of arrival times to the time of stimulation. The directional dependence of the conduction velocity was consistent with previous reports: at 1 mA, longitudinal (0°) and transverse (90°) velocities were 0.60 ± 0.03 and 0.29 ± 0.02 m/sec, respectively. At 7 mA, the longitudinal velocity was 0.75 ± 0.05 m/sec, whereas there was no significant change in the transverse velocity. In contrast to conduction velocity, the virtual cathode was smallest in the longitudinal orientation and largest between 45° and 60°. Virtual cathode size was dependent on both orientation and stimulus strength: at 0°, the virtual cathode was small (~ 1 mm) and relatively constant over the range of 1-7 mA; at oblique orientations (45°-90°), it displayed a roughly logarithmic dependence on stimulus strength, approximately 1 mm at 1 mA and approximately 3 mm at 7 mA. The bidomain, anisotropic model reproduced both the stimulus strength and the fiber-orientation dependence of the virtual cathode geometry when the intracellular and extracellular anisotropies were 10:1 and 4:1, respectively, but not when the two anisotropies were equal. We suggest that the virtual cathode provides a direct measure of the determinants of cardiac activation; its complex geometry appears to reflect the bidomain, anisotropic nature of cardiac muscle.