Phenotypes were available on 4 production traits recorded for up to 196,613 broiler chickens. Heritabilities ranged from 0.22 to 0.49. Among all phenotyped birds, 15,723 were genotyped for 39,102 segregating SNP. Traditional and genomic evaluations were run in a multiple-trait model. Single-step genomic BLUP (ssGBLUP) was used for genomic evaluations with 3 different reference sets including only males (4648), only females (8100), and both sexes (12,748). Realized accuracy of genomic EBV (GEBV) was used to evaluate the inclusion of genotypes for different reference sets on predictive ability of young genotyped males (1501), females (1474), and both sexes (2975). Using male genotypes as reference, the average increase in accuracy of GEBV over EBV for males and females was 12 and 1 percentage point, respectively. When the reference population included only female genotypes, the increase for males and females was 1 and 18 percentage points, respectively. Using genotypes on both sexes as reference increased accuracies by 19 points for males and 20 points for females. Adding genotypes for females without phenotypes did not improve predictions. For one trait, EBV and GEBV accuracies for females were much lower than for males. For another trait with similar heritability, both accuracies were higher, and females had higher accuracy than males. For validation animals GEBV ≈ w₁PA + w₂DGV, where PA is parent average, DGV is genomic prediction, and the w terms are the weights for each component. When the number of genotyped animals is high, the highest weight is for w₂. When an animal is genotyped, the increase in accuracy comes mainly from the DGV portion of GEBV and marginally from improved PA. For non-genotyped animals there is no improvement in accuracy due to DGV. Accuracies for animals of one sex increase with genotypes of the other sex when that sex has independent phenotypic information and when the evaluation methodology avoids double counting. Realized accuracies are biased down by selection, and analysis of realized accuracies can reveal different selection pressure for traits and sex.