Traditional marker assisted selection uses two steps; mapping QTLs followed by selection using markers linked to selected QTLs. A disadvantage of this approach is that the threshold for detecting a QTL is arbitrary and can affect the resulting selection. Genomic selection overcomes this disadvantage by directly estimating all the marker effects across the genome to produce genomic estimated breeding values (GEBV). In this study, 200 breeding lines were evaluated for yield and heading date in two locations in Minnesota (St.Paul and Crookston). These lines were genotyped with 1,536 SNP markers (BOPA1) for all seven barley chromosomes. Ridge regression (RR) is used to train a model for genomic selection for these traits. We are using leave-one family out cross-validation to evaluate model accuracy using the 200 breeding lines. We also assessed model accuracy on sets of lines developed over ten years period following this training data set to determine the effect of time on model accuracy.