The objective was to assess by simulation the efficacy of population structure analysis in plant breeding. Twelve populations and 300 inbred lines were simulated and genotyped using 100 microsatellite loci. The experimental material included populations with and without admixture, ancestry relationship and linkage disequilibrium, and with distinct levels of genetic differentiation and effective sizes. The analyses were performed using Structure software and employed all available models. For all the group number (K) tested, for both populations and inbred lines, the admixture model with correlated allelic frequencies provided the highest value for the logarithm of the marginal likelihood. Fitting appropriate model and using adequate sample size for individuals and markers, Structure was effective in identifying the correct population structure, migrants and individuals with genome from distinct populations. The linkage model did not result in an improvement in clustering relative to the admixture model with correlated allelic frequencies. The inclusion of prior information did not change the results; for some K values the analyses showed slight higher values of the marginal likelihood. The reduction in the number of individuals and markers negatively affected the results. There was a high variation in the most probable K value between the evaluated methods.