The availability of the 54k goat chip has allowed the genotyping of 825 males from the French goat selection scheme for the two breeds Alpine and Saanen. In this species, genomic selection could improve annual genetic gain by reducing the length of the sire-son pathway (5.5 years in 2011).This would require that accuracy of genomic breeding value of young males (without phenotype) is at least equal to the parent average one. Moreover, the quality of prediction being sensitive to the size of the reference population (sires already genotyped and phenotyped), a multi-breed (combining Alpine and Saanen performances) approach could be considered. The first part of the study focuses on genomic structure of the reference populations of these two breeds that may impact on expected accuracy of genomic predictions. Persistence of linkage disequilibrium (LD) phases between Alpine and Saanen, 0.58 at 60kb, was the same as the one find between Manech and Lacaune dairy sheep breeds. Moreover, a low correlation (68%) between allele frequencies in Saanen and Alpine breeds and an analysis in principal component based on genotypes have shown that Alpine and Saanen have to be considered as two different breeds. Genetic correlation between the two breeds estimated on SNP data ranged between 0.45 for milk yield and 0.76 for udder floor position. Genomic selection was tested using a single step approach applied to a GBLUP model in the multi-breed or within-breed for production and udder type traits. The prediction quality was assessed by cross validation and the accuracy of genomic breeding values of candidates was compared to the parent average one. The prediction quality using multi-breed or within-breed model was similar, between 0.34 for fore udder in Alpine breed to 0.73 for protein content for Saanen breed. Within-breed, these values were higher in Saanen due to an higher inbreeding level, and similar as the one obtained in Lacaune dairy sheep with a reference population of 2 185 rams. Genomic breeding value accuracy of candidates exceeds parent average accuracy from 10% for protein yield in Alpine breed to 14% for protein content in Saanen breed. Clearly the single step GBLUP method provides opportunities for genomic selection in French goats’ population.