Research over the last few decades has led to contradictory insights on the genetic architecture and evolution of quantitative traits. Thus, key questions about the relevance of large-effect genes identified in some studies for understanding adaptation and evolution are yet to be resolved. To address these issues, models of quantitative trait variation need to incorporate the contribution of mutations at multiple sites within genes and the potential for recombination between these sites. In addition, predictable patterns of genetic architecture may be specific to particular combinations of evolutionary forces. We are using quantitative genetic simulations to test whether evolutionary genetic models incorporating different modes of selection can explain discordant results from different study systems. In contrast with other modeling efforts, we incorporate both the cumulative effects of multiple mutations in the same gene and realistic rates of intragenic recombination in our simulations. Preliminary results suggest that sharply contrasting genetic architectures may arise in a single population under stabilizing selection versus populations under divergent selection for contrasting phenotypic optima. Unlike the highly polygenic architecture generated under stabilizing selection, divergent alleles at only a few loci soon come to explain most of the differentiation between populations under scenarios examined so far. However, large allelic effects tend to arise from the cumulative effect of many mutations in multiple potentially-recombining segments of the same locus. These results appear to support previous suggestions that micromutational models and differentiation at large-effect loci are not mutually exclusive. We discuss extension of our modeling to test whether “stepping stone” differentiation along environmental gradients produces similar or different patterns, and address insights into the roles of preexisting variation vs. novel mutations in adaptive divergence.