A simulation approach was used to study bias and variability of density estimates of fish using the transect and point count underwater visual census methods. Three experiments were conducted to examine the effects of fish density, sampling effort, and the speed of fish in relation to the observer. Fish density and sampling effort did not significantly bias estimates of fish density using either census method, and variation was a function of the area sampled with both methods. The speed at which fish approached the diver caused appreciable bias with the transect method but not with the point count method, because of underlying assumptions about how the two methods were implemented. Performance of methods was quantified with the root mean squared error RMSE (combined measure of bias and variability) and was dependent on the ratio of sampling times per dive for each method. From assumed sampling times for the point count and transect methods, the point count method performed better than the transect method, but different results could be obtained under different sampling protocols. Nevertheless, the simulation approach offers an efficient means to evaluate sampling methods in conjunction with actual field experiments.