In a case-parent trio study we collect genotypes on affected children and their parents. Information may also be collected on the child's environmental exposures. The design permits estimation and testing of genetic effects and gene-by-environment interaction. Inference of genetic effects is robust to population structure, but when genotypes are measured at a non-causal test locus, population stratification can create spurious interaction. That is, the exposure can appear to modify the disease risk of genotypes at the test locus without actually modifying the disease risk of genotypes at the causal locus. We review previous methods to reduce bias from population stratification and propose a new method in which we adjust the risk model by principal components computed from a genome-wide panel of markers. The method is illustrated on simulated data and on data from a study of genetic modifiers of exposures known to affect the risk of cleft palate.