Improved genetic mapping approaches are key to a range of biological studies, including investigations of the genetic basis of adaptive trait evolution. Here, we develop a next generation approach to bulk segregant analysis (BSA) in Drosophila, using a simulation-based inference framework to deal with overlapping quantitative trait locus (QTL) peaks. Our simulations suggest that BSA yields consistently stronger QTL signals than introgression mapping. We apply this method to traits that differ adaptively between natural D. melanogaster populations. These traits include body and wing size in a highland Ethiopian population, and the parallel evolution of melanic pigmentation and ethanol tolerance in three populations each. QTLs of moderate to large effect are readily detectable for each trait, but these often vary among strains from the same population. Examining genetic variation within QTL peaks, we find evidence that local adaptation occurred via selection on standing genetic variation, producing “soft sweep” patterns such as isolated variants with strong population frequency differences. These results begin to crystalize a general model for the genetic architecture of adaptive trait evolution in Drosophila. The BSA mapping method introduced here may prove valuable in many studies in Drosophila and in other experimentally tractable organisms.