Bacterial strains are currently grouped into species based on overall genomic similarity and sharing of phenotypes deemed ecologically important. Many believe this polyphasic taxonomy is in need of revision because it lacks grounding in evolutionary theory, and boundaries between species are arbitrary. Recent taxonomy efforts using multilocus sequence typing (MLST) data are based on the identification of distinct phylogenetic clusters. However, these approaches face the problem of deciding the phylogenetic level at which clusters are representative of evolutionary or taxonomically distinct units. In this review, I propose classifying two phylogenetic clusters as separate species only when they have statistically significantly diverged as a result of adaptive evolution. More than a method for classification, the concept of adaptive divergence can be used in a 'reverse ecology' approach to identify lineages that are in the process of speciation or genes involved in initial adaptive divergence.