Genomic prediction has proven its efficiency in numerous animal and plant species. In this study, we used diverse lucerne varieties and populations to test the predicting ability of genomic prediction. Several parameters, such as the number of markers, the population size and the addition of QTL effects, were tested for their effect on the quality of prediction. Based on a large number of SNPs (227 K) obtained by GBS and phenotypes observed in different locations, our results showed a good quality of predicting ability for dry matter yield, ADF (acid detergent fiber) and protein content, especially with a large training population size (around 0.6). The predicting ability is improved by the integration of QTL information directly in the model (above 0.8). A reduction of number of markers (less than 100K) did not alter much the predictive ability. Our results show an accurate prediction of the phenotype of populations via genomic prediction models that could speed up the creation of new lucerne varieties.