Elevated levels of genetic and environmental variance are often expressed in extreme habitats, but it is unclear whether it is because of their novelty, or their low quality. Using a half-sib-full-sib design, I measured the components of phenotypic variance in life-history traits (developmental time, emergence weight, growth rate, lifetime fecundity and male life span), and heritability of larval survival, expressed by the cowpea weevil, Callosobruchus maculatus, on the usual and three novel host species. Because the novel hosts differed in quality, I was to some extent able to distinguish between the effects of host novelty and quality. For all traits except larval survival the among-sire variance component was consistently higher on the usual host, although this result was statistically significant only when the estimates for different traits were standardized and pooled. It contradicts the predictions of evolutionary theories. In turn the dam component for these traits tended to be smallest on the usual host, but this pattern was less pronounced. The within-family variance component decreased with improving performance. This result supports the hypothesis that canalization of development against environmental fluctuations is less efficient in poor habitats. I found no effects of host novelty on environmental variance. Heritability of larval survival was lowest on the usual host, in accord with the hypothesis that predicted higher expression of genetic variance in novel environments.