Single-step genomic BLUP (ssGBLUP) is a method that combines all sources of information in a single analysis to compute genomic EBV (GEBV). For single-step genome-wide association studies (ssGWAS), GEBV are backsolved to SNP effects, and those effects are converted to proportion of explained additive genetic variance. Thus far, no formal framework for hypothesis test is currently present in ssGWAS from the BLUPF90 software suite. Our objective was to implement P-values for ssGWAS and to apply the method to a large dairy cattle population. P-values were obtained based on the prediction error (co)variance for SNP, which uses the inverse of the coefficient matrix for genotyped animals and formulas to compute SNP effects. Six steps are needed for the calculation of P-values: 1) factorize and invert the LHS of ssGBLUP; 2) solve MME using sparse Cholesky factor; 3) extract the LHS⁻¹ for genotyped animals; 4) backsolve GEBV to SNP effects; 5) obtain the prediction error covariance for SNP effects; 6) calculate P-values using the cumulative standard normal function of SNP effect divided by standard deviation of SNP effect. The US Holstein data used in this study consisted of almost 800k udder depth records for 500k cows. Pedigree information was available for 1.3M animals, of which 8,802 sires were genotyped. The model contained the same effects as the official model used for linear type trait evaluation in the US; however, in a single-trait setup. Computation of P required 20Gb of memory and no inflation was observed. The SNP passing the Bonferroni threshold of 6.1 in the −log10 scale were the same as those that explained the highest proportion of additive genetic variance. The exact P-value for ssGWAS is a very general and efficient strategy for QTL detection and test. It can be used in complex data sets such as the ones used in animal breeding, where only a proportion of pedigreed animals are genotyped. The BLUPF90 software suite is now equipped with the P-value calculation tool.