Many methods have been proposed for the implementation of genomic selection in the beef industry. The structure of the beef industry is sufficiently different from the dairy industry to warrant consideration of smaller, targeted panels of SNPs which are more cost-effective for the estimation of genomic breeding values for commercial producers. We explore methodologies to construct training and validation populations in admixed purebred (Angus n=651; Hereford n=1,095) and crossbred (Limousin n=283; Charolais n=695 and Simmental n=516) populations from the Carcass Merit Project. Genomic selection models have been built using 40,645 SNPs from the BovineSNP50 chip within GenSel, a Bayesian modeling program developed at Iowa State University. Emphasis was given to fitting models such as BayesCπ, which estimates the proportion of markers that do not affect a trait from the data and allows identification of the most important SNPs for prediction of genetic merit in a trait. Incorporation of information on genetic distance between breeds into selection of animals for training and validation populations did not result in accuracies significantly higher than randomly allocating animals to training and validation populations. Preliminary results obtained by randomly allocating animals to training and validation show accuracies from internal validation ranging from 0.59 to 0.86. Additional external validations will be performed for Warner-Bratzler Shear Force to help ascertain the amount of linkage disequilibrium included in the predictions as well as how well these predictions perform in a population separated from training by approximately two generations.