In pig breeding, the final product is a crossbred (CB) pig. Selection of purebreds (PB) for CB performance, thus, may result in greater genetic progress. The aim of this study was to assess empirically the accuracy of genomic breeding values (GEBV) of PB animals for CB performance. Phenotypes and genotypes were available from sows of three pig populations: 1,668 Dutch Landrace (DL), 2,003 Large White (LW) and 914 CB, an F1 cross between DL and LW. Two traits were analyzed: gestation length (GLE) and total number of piglets born (TNB). Scenarios studied were prediction within population and prediction of PB genetic merit based on CB animals in the training set. Within-population GEBVs were validated by comparison with corrected phenotypes. PB genetic merit for CB performance was validated by comparison to the PB animals’ deregressed breeding value (DEBV) estimated from pedigree BLUP with CB phenotypes. GEBVs were estimated with GBLUP using ASReml software. Accuracy was computed as the correlation between the GEBV and the corrected phenotype/DEBV of the validation set animals divided by the square root of the heritability/reliability of the validation population. Accuracy of within-population predictions ranged from 0.36 to 0.66 across the traits and training sets. Accuracy for prediction of PB genetic merit for CB performance for GLE was 0.27 in DL pigs and 0.23 in LW pigs. For TNB accuracy was 0.11 for DL and 0.22 for LW. In sum, an encouraging level of predictive ability was observed for genomic prediction of PB genetic merit for CB performance. Our results show that the prediction of CB performance in PB lines is promising for the traits studied.