Joseph Pickrell and Jonathan Pritchard report a new statistical model for demographic inference from genome-wide allele frequency data sets (PLoS Genet, 8, e1002967, 2012). Implemented in the freely available software TreeMix (http://treemix.googlecode.com/), their approach involves building a population tree, finding populations that show poor fit to the tree model and modeling migration events to improve the fit within a graph-based model. They tested their method in simulations as well as on a human data set including genome-wide data from 53 modern and 2 archaic populations. They inferred a maximum-likelihood tree from the human data set that recapitulates known population relationships and explains 98.8% of the variance in relatedness between populations. By sequentially adding the ten migration events that are most consistent with known events, they increased the explained variance to 99.8%. The authors also applied their approach to a canine genome-wide data set of 82 dog breeds or wild canids. They found that a significant amount of gene flow occurred between breeds during dog domestication and again demonstrate the ability to increase the explained variance in relatedness between dog breeds by sequentially adding migration events. For both humans and dogs, the authors infer many migration events, including previously known and new relationships between populations.