This study was aimed at understanding the genetics of growth in mice lines divergently selected body weight for 50 generations and to investigate the linearity or non-linearity of offspring-parent regression obtained from unselected, random bred mice lines. A large body of data comprising 1090 offspring (family mean) - parent pairs was used to investigate the relationship between offspring and parents. Linear and polynomial models were fitted for the regression of 6-week body weight in offspring on one or both parents. Regression analysis of offspring mean on mid-parent showed that heritability of 6-week body weight was similar to the estimates from simultaneously regressing offspring on both parents provided that the records were standardised within parent-sex category. Regression of offspring family means (sons and daughters) on sire's weights were somewhat non-linear, while offspring-dam regressions were generally linear. Multivariate regression analyses combining both parents gave similar results compared to results of offspring-one parent regression. The second study used data from 30 generations of the P6-lines divergently selected (within family) for high and low 10-week body weight (generation 21-50). These lines were previously selected for 20 generations based on an index of lean mass (P-lines), and crossed to form the P6-lines. The mean divergence in 10-week body weight at generation 21 was 13.3 g increasing to 32.2g at generation 50. The overall divergence relative to the base population (i.e. the P-lines) was 6.3σp. The low line (L) appeared to plateau at generation 40 and the plateau was associated with reduction in selection differentials. When response was considered separately for each sex in the P6-lines, females of the high line (H) responded more to upward selection than in the males, while males responded more than females in the L line. Sexual dimorphism (ratio of male/female mean) was consistently higher in the L line. The genetic correlation between male and female 10-week body weight estimated by Restricted Maximum Likelihood (REML) was 0.84±0.03.