When alternative models are used for genome-wide association studies (GWAS), not a single one of them has been recognized as universally best across all traits. One explanation is that different traits may have different genomic architecture characterized by different distributions for marker effects. Further, it is known that results from BayesA and BayesB models can be heavily influenced by the value of hyperparameters assumed known, namely pi, the scale factor and degrees of freedom. Our objective was to use a robust Bayesian regression model for GWAS that treated these hyperparameters as unknown and apply this model to 3 growth traits in Brangus (3/8 Brahman × 5/8 Angus) heifers. Genotypes for each heifer were obtained from BovineSNP50 Infinium beadchips. Phenotypes included data on birth weight (BW), weaning weight (WW) and yearling weight (YW) from 830 individuals from 67 sires. Simultaneous association of all SNP for each of BW, WW and YW were tested in a robust model that treated SNP effects as random. Fixed effects included cohort groups as class effects (defined for animals with the same calving season, location, and trait contemporary group) and dam age (in years) as classes. Several genomic locations were associated with variation in these growth traits in heifers. The top 10 regions by SNP effects were found on chromosomes 1, 2, 5, 6, 14 and 26 for BW, chromosomes 1, 2, 3, 5, 6, 16 and 28 for WW and chromosomes 5, 6, 7, 11, 12, 14, 16, 18, 20 and 29 for YW. Results confirm many previously reported regions associated with variation in these growth traits in both taurine and indicine cattle breeds but also included new associations. Results demonstrated the utility of regression models with unknown hyperparameters of pi, scale and degrees of freedom in GWAS of growth traits in Brangus heifers.