Performance of purebred parents can be a poor predictor of performance of their crossbred descendants. However, in crossbred populations linkage disequilibrium may not be restricted to markers that are tightly linked to the QTL and the effects of SNPs may be breed specific. Both these problems can be addressed by using a model with breed-specific SNP effects. To investigate this idea, we used a model with effects of SNP genotypes (model 1) and a model with breed-specific effects of SNP alleles (model 2) to predict purebred breeding values using F₁ data. Three scenarios were considered. In the first two, pure breeds were assumed to have a common origin either 50 or 550 generations ago. In the third scenario, the two breeds did not have a common origin. In all these scenarios, the two breeds were used to generate an F₁ with 1,000 individuals. Trait phenotypic values controlled by 12 segregating QTL and with a heritability of 0.30 were simulated for the F₁ individuals. Further, 500 segregating markers on a chromosome of 1 Morgan were chosen for analysis in the F₁s and in the validation population of purebred. A Bayesian method (Bayes-B) was used to estimate the SNP effects. The accuracy of the predictions was between 0.789 and 0.718. However, the accuracy was lower when the populations were more separate and model 2 gave values slightly higher than model 1. These results suggest that crossbred data could be used to evaluate purebreds and breed specific models could give better results.