Cave organisms occupy a special place in evolutionary biology because convergent morphologies of many species demonstrate repeatability in evolution even as they obscure phylogenetic relationships. The origin of specialized cave-dwelling species also raises the issue of the relative importance of isolation vs. natural selection in speciation. Two alternative hypotheses describe the origin of subterranean species. The 'climate-relict' model proposes allopatric speciation after populations of cold-adapted species become stranded in caves due to climate change. The 'adaptive-shift' model proposes parapatric speciation driven by divergent selection between subterranean and surface habitats. Our study of the Tennessee cave salamander complex shows that the three nominal forms (Gyrinophilus palleucus palleucus, G. p. necturoides, and G. gulolineatus) arose recently and are genealogically nested within the epigean (surface-dwelling) species, G. porphyriticus. Short branch lengths and discordant gene trees were consistent with a complex history involving gene flow between diverging forms. Results of coalescent-based analysis of the distribution of haplotypes among groups reject the allopatric speciation model and support continuous or recurrent genetic exchange during divergence. These results strongly favour the hypothesis that Tennessee cave salamanders originated from spring salamanders via divergence with gene flow.