Expected genetic response is proportional to the heritability of the trait, and this parameter is considered inherent of a specific trait in a particular population. However, models assuming heterogeneity in residual variance lead to different estimates of heritability across combinations of systematic (environmental) effects. Modifying the residual variance of the birth weight by artificial selection was shown to be feasible in a divergent selection experiment in mice. The objectives of this work were to 1) estimate the evolution of the heritability of birth weight in mice in the mentioned experiment, and 2) estimate different heritability regarding systematic effects. Data came from eleven generations of a divergent selection experiment to modify the residual variability of birth weight in mice. A total of 15,431 birth weight records from 959 females and 1,641 litters in combination with 14,786 pedigree records were used. The model used for analysis included generation, litter size, sex, and parity number as systematic effects. Each record of birth weight was assigned to the mother of the pup in the model which assumes that the residual variance is heterogeneous and partially under genetic control. Differences in heritability between lines reached values of 0.06 in the last generations. Choosing the most extreme values of systematic effects, the birth weight heritability ranged from 0.04 to 0.22. From these results, the possibility of modulating the heritability for this trait could be explored in 1 of 2 ways: selecting to decrease the residual variability, or choosing the specific levels of the systematic effects.