Interest in genetic improvement of carcass and tenderness traits of beef cattle using genome-based selection (GS) and marker-assisted management programs is increasing. The success of such a program depends on the presence of linkage disequilibrium between the observed markers and the underlying QTLs as well as on the relationship between the discovery, validation and target populations. For molecular breeding values (MBVs) predicted for a target population using SNP markers, reliabilities of these MBVs can be obtained from validation analyses conducted in an independent population distinct from the discovery set. The objective of this study was to test MBVs predicted for carcass and tenderness traits of beef cattle in a Canadian-based validation population that is largely independent of a US-based discovery set. The discovery dataset comprised of genotypes and phenotypes from > 2,900 multi-breed beef cattle while the validation population consisted of 802 crossbred feeder heifers and steers. A bivariate animal model that fitted actual phenotype and MBV was utilised for validation analyses. The reliability of MBVs was defined as square of the genetic correlation (R(2)g) that represents the proportion of the additive genetic variance explained by the SNP markers. Several scenarios involving different starting marker panels (384, 3K, 7K and 50K) and different sets of SNPs selected to compute MBVs (50, 100, 200, 375, 400, 600 and 800) were investigated. Validation results showed that the most reliable MBV (R(2)g) were 0.34 for HCW, 0.36 for back fat thickness, 0.28 for rib eye area, 0.30 for marbling score, 0.25 for yield grade and 0.38 for Warner-Bratzler shear force across the different scenarios explored. The results indicate that smaller SNP panels can be developed for use in genetic improvement of beef carcass and tenderness traits in order to exploit GS benefits.