Wright first introduced the idea that random genetic drift and classical mass-action selection might combine in such a way as to allow populations to find the highest peak in complicated adaptive surfaces. His theory assumes large but structured populations, in which mating is spatially local. If gene flow is sufficiently low, and the subpopulations (demes) are small enough, they will be subject to genetic drift. Distant demes drift independently, allowing many independent searches of the adaptive surface to take place. A deme that has shifted to a higher peak can, by emigration, cause the rest of the demes to shift to the higher peak. The probability of this shift depends on the migration rate. Previous studies have concluded that very little migration is necessary to effect the shift in adaptive peaks that characterizes the last phase of Wright's Shifting Balance Process (SBP). Here we present the results of a computer study that investigates the roles of dispersal distance, the degree of epistasis in the fitness surface, and recombination on the shifting balance process. In particular, we measure their effect on the population's mean fitness. We show that over a range of dispersal distances the advantage of the SBP is a monotonically increasing function of the amount of epistasis. Our results show that the extent of dispersal that results in the greatest effect of the SBP in increasing mean fitness depends on the extent of epistasis. Finally, for low levels of epistasis, higher recombination performs better, while for intermediate levels, lower recombination results in a greater advantage of the SBP.