Selective breeding programs have been initiated worldwide for the Pacific oyster Crassostrea gigas to improve economically important traits such as growth and disease resistance. The emergence of genomic tools has allowed novel perspectives for using genomic selection (GS) in mixed-family breeding designs, which are cheaper and easier to develop than classical breeding schemes. In this study, we evaluated the potential of GS for different growth-related and shell colour traits in two independent commercially selected populations (P1 and P2), based on mixed-family designs. From ≈14.5K informative SNPs genotyped with the bi-species Axiom Affymetrix 57K oyster array, ≈12.5K were mapped on the reference genome. A strong heterogeneity of marker density between and within chromosomes was reported, with a low linkage disequilibrium (below 0.1 at 0.1 Mb) between pairs of SNPs. The within-population structure was homogenous for each population, with effective sizes of 107 for P1 and 76 for P2. Heritability was estimated for each trait and population and ranged from 0.08 ± 0.04 (for mean darkness intensity in P1) to 0.56 ± 0.08 (for the mean upper valve b* value in P2) for a pedigree-based model and from 0.04 ± 0.02 (for mean darkness intensity in P1) to 0.69 ± 0.04 (for the mean darkness intensity in P2) for a genomic-based model. Growth-related traits were generally highly genetically and positively correlated with each other, but weakly correlated with colour traits. Accuracy of prediction was generally higher with the genomic model (GBLUP) than with the classical BLUP model, with a maximum gain of accuracy (from 0.38 to 0.66) for flesh weight adjusted by total weight in P2. Accuracy of breeding values was slightly higher for colour traits for P2, with higher heritability estimates. Overall, our results indicate that GS has a good potential to be implemented in mixed-family breeding programs in a shellfish such as C. gigas.