Mixture models provide a powerful way of teasing apart population structure from genetic data, as exemplified by the program STRUCTURE. However, choosing the number of sub-populations (denoted K) is a more difficult problem as this requires evaluation of the *evidence* of each model. In contrast to other programs, *rmaverick* uses thermodynamic integration to arrive at precise and accurate estimates of the evidence for each *K*, allowing the user to choose appropriate values of this parameter, and also to compare between different evolutionary models (e.g. with/without admixture).