Data from the Spanish Holstein-Friesian milk recording system were used to investigate sources of heterogeneity of genetic and residual variation and to evaluate the impact in the genetic evaluation. Factors such as period of time, herd level of production, geographical region, herd size and year of calving were found to be associated with heterogeneity of genetic and residual variation. This stratified the 7664 herd-year of calving into 395 strata defined by combinations of levels of these factors. In order to improve the accuracy of estimates of strata dispersion a mixed model structure in the log-linear model was considered which included the single factors acting additively as fixed effect plus the random third order interaction for residual variance and the single factor acting additively as fixed effect for sire variance. The average value for heritabilities estimated was 0.26, with maximum and minimum values being 0.52 and 0.11, respectively. The 25 and 75 percentiles were 0.20 and 0.32, respectively. Genetic evaluations were compared using three alternative models differed in the dispersion structure Homogeneous genetic and residual variance, homogeneous genetic and residual variance after data standardization and the heterogeneous variance model with the current genetic and residual variance estimated. Assuming a homogeneous variance model produced an overdispersion with respect to the expected dispersion of both yield deviations and predicted genetic values in large variance environments and the contrary was observed in low variance environments. Precorrection by phenotypic standardization seems to correct for the phenotypic adequately but dispersion of predicted genetic values of animal in environments with large heritability was underestimated with respect to the heterogeneous variance model.