This paper presents a statistical model for inferring population splits and mixtures from genome-wide allele frequency data. The model uses a graph-based approach to represent the relationships between populations, allowing for both population splits and gene flow. The method was applied to 55 human populations and 82 dog breeds and wild canids. In humans, the model identified that Cambodians trace approximately 16% of their ancestry to a population ancestral to other East Asian populations. In dogs, it showed that the boxer and basenji breeds trace a considerable fraction of their ancestry to wolves, and that East Asian toy breeds result from admixture between modern and ancient breeds. The model, called TreeMix, is available as open-source software. The study demonstrates that a simple bifurcating tree is insufficient to describe population history, and that migration events are common. The model was tested on simulated data and applied to real data, showing high accuracy in inferring population relationships and identifying admixture events. The results suggest that gene flow has played a significant role in the history of both humans and dogs. The model is computationally efficient and can be applied to large datasets. The study highlights the importance of considering gene flow in population genetic analyses and provides a framework for understanding the complex demographic history of species.This paper presents a statistical model for inferring population splits and mixtures from genome-wide allele frequency data. The model uses a graph-based approach to represent the relationships between populations, allowing for both population splits and gene flow. The method was applied to 55 human populations and 82 dog breeds and wild canids. In humans, the model identified that Cambodians trace approximately 16% of their ancestry to a population ancestral to other East Asian populations. In dogs, it showed that the boxer and basenji breeds trace a considerable fraction of their ancestry to wolves, and that East Asian toy breeds result from admixture between modern and ancient breeds. The model, called TreeMix, is available as open-source software. The study demonstrates that a simple bifurcating tree is insufficient to describe population history, and that migration events are common. The model was tested on simulated data and applied to real data, showing high accuracy in inferring population relationships and identifying admixture events. The results suggest that gene flow has played a significant role in the history of both humans and dogs. The model is computationally efficient and can be applied to large datasets. The study highlights the importance of considering gene flow in population genetic analyses and provides a framework for understanding the complex demographic history of species.