The genetic architectures of four morphometric traits in the housefly, Musca domestica L., were analysed using a paternal half-sib design on a large outbred base population and on two populations each established with two pairs of flies from the base population. By including parents with the offspring in the analysis, phenotypic variance was subdivided into components for additive, dominance, additive-by-additive epistasis and environmental variation. In relation to the nonbottlenecked base population, narrow sense hertitability was lower in the bottlenecked populations, largely because of an increase in the level of the environmental component of variance. Averaged over the four traits in the nonbottlenecked base population, nonadditive components contributed nearly 60 per cent of the phenotypic variance, with more than half of this being attributable to epistasis. Two of the four traits showed reduced additive genetic variance in the bottlenecked lines, whereas two of the traits showed increased additive genetic variance. A major difference in the genetic architecture between traits showing these two disparate bottleneck effects was the presence of epistasis in the nonbottlenecked base population for the traits which showed an increase in the additive component. Overall, there was a significant positive relationship between the initial percentage contribution of epistasis in the base population and the gain in the additive component in the bottleneck lines, as predicted by models incorporating additive-by-additive epistasis.