Data from 23 generations of mice selected for increased and reduced appetite were analysed by Restricted Maximum Likelihood fitting an animal model with litters as additional random effects. Traits considered were food intake between 4 and 6 weeks of age adjusted for 4-week body weight (AFI), the selection criterion, and body weight at 6 weeks (6WW). Selection was carried out within families. A high and a low selection line and a control were maintained in each of three replicates. Analyses were performed for each replicate separately taking subsets of the data spanning different numbers of generations. Overall estimates of heritabilities were 0.15 for AFI, which agreed well with realized heritability estimates, and 0.42 for 6WW. The litter variance, expressed as a proportion of the phenotypic variance, was 0.21 for both traits, yielding intraclass correlations of full-sibs of 0.29 and 0.42, respectively. Similar results were obtained for variances of each trait using univariate and multivariate analyses. From the latter, estimates of correlations between the two traits were 0.46 for additive genetic, -0.19 for litter and 0.31 for residual effects, resulting in a phenotypic correlation of 0.23. Analyses of data from generations 2-7, 8-13 and 14-23 separately showed a marked decrease in genetic variance and heritability in later generations for both traits. Heritabilities of AFI, for instance, were 0.24, 0.10 and 0.07, respectively. These changes could not be attributed to the effects of inbreeding or of selection in an infinitesimal model and suggested that some change in variance due to change in gene frequency had occurred during the course of the experiment.