Simulations of the electrical activity during excitation were performed in an anatomically based model of the human ventricular conduction system. Each of the 33,000 elements of this model represented a unit bundle of Purkinje or atrioventricular nodal tissue. The Ebihara-Johnson model for sodium defined the active membrane characteristics. Using a combination of new and existing modeling techniques, simulations of excitation were completed in approximately 5 min CPU time on an IBM 3090 at the Cornell National Supercomputer Facility. Activation times at sites in the model were compared to experimental measurements for the excitation of the ventricular myocardium on the endocardial surface. These "literature-based" times were estimated from a number of reported human heart mapping studies. Initially, the times fit poorly. The major factor for the discrepancy was the conduction velocities of the elements, which were a result of the physical and electrical parameters derived from a review of histologic and electrical properties studies. In addition, there was a latency between activation of the system in the left ventricle of the model and that in the right ventricle when compared to the experimental work. When the times were scaled to adjust for the conduction velocity and ventricular latency effects, the match between the simulation and literature-based times was much improved. Quantitative comparison between normalized times resulted in correlation coefficients CCF = 0.76 for the right ventricle and CCF = 0.64 for the left ventricle.