A multi-breed genome wide association study (GWAS) can potentially improve QTL mapping precision and detection power. Alternatively to a multi-breed GWAS, meta-analysis can combine within breed GWAS results. Our objective was to compare within breed GWAS, multi-breed GWAS and meta-analysis of within breed GWAS results. Imputed sequences and deregressed proofs for milk, fat and protein yield of 16,031 bulls of five French and Danish dairy cattle breeds were used for the analyses. GWAS were performed within each breed, combining French and Danish Holstein, combining Jersey, Montbéliarde, Normande and Danish Red, and combining all breeds. Within breed GWAS results were combined using three different meta-analysis models. The multi-breed GWAS resulted in more distinct peaks by increasing the p-values of some variants and decreasing the p-values of others. For some QTL not segregating in Holstein, combining all breeds except Holstein was useful, because they were overshadowed by larger QTL segregating in Holstein when all breeds were combined. The meta-analysis gave results similar to the multi-breed GWAS and can be used as an alternative. The results obtained by the weighted Z-score model were closest to those of the multi-breed GWAS.