A method based on the animal model is described which allows the estimation of continuous changes in variance components over time using restricted maximum likelihood (REML). The method was applied to the analysis of a selection experiment in which a foundation population formed from a cross between two inbred strains of mice (C57BL/6J and DBA/2J) was divergently selected for 6 week body weight over 20 generations. The analysis suggested that there was an increase in phenotypic variance of about 50% in the low selected lines over the course of the experiment which was attributed to increases in the environmental and additive variance components. Variance changes in the High selected lines were generally smaller than in the Low lines, although there was an estimated 20% increase in the environmental variance. Simple models to explain these effects involving dominance, linkage and epistasis were explored. Testing which of these was responsible for the variance changes noted in this experiment (if any) is difficult, although the epistasis and dominance models require less stringent conditions than the linkage model, and the dominance model is supported by evidence of heterosis in the F1.