The modes of speciation that are thought to have contributed most to the generation of biodiversity require population di¡erentiation as the initial stage in the speciation process. Consequently, a complete understanding of the mechanisms of speciation requires that the process be examined not just after speciation is complete, or nearly so, but also much earlier. Because reproductive isolation de¢nes biological species, and it evolves slowly, study of the process may require a prohibitive span of time. Even if speciation could be observed directly, selection of populations in the process of speciation is typically di¤cult or impossible, because those that will ultimately undergo speciation cannot be distinguished from those that will di¡er-entiate but never assume the status of new biological species. One means of circumventing this problem is to study speciation in taxa comprising several sibling species, at least one of which exhibits extensive population di¡erentiation. We illustrate this approach by exploring patterns of population variation in the postglacial radiation of the threespine stickleback, Gasterosteus aculeatus. We focus on lacustrine populations and species within this complex, demonstrating parallel axes of divergence within populations, among populations and among species. The pattern that emerges is one of parallel relationships between phenotype and ¢tness at all three hierarchical levels, a pattern that facilitates exploration of the causes and consequences of speciation and secondary contact. A second outcome of this exploration is the observation that speciation can be the consequence of a cascade of e¡ects, beginning with selection on trophic or other characteristics that in turn force the evolution of other population characteristics that precipitate speciation. Neither of these conclusions could have been reached without comparative studies of wild populations at several hierarchical levels, a conclusion reinforced by a brief survey of similar e¡orts to elucidate the process of speciation. We address the issues most likely to be resolved using this approach, and suggest that comparisons of natural variation within taxa at several hierarchical levels may substantially increase our understanding of the speciation process.