The spurious interaction arises from imperfect LD between the marker and causal variant, causing a misclassification error that differs between cases and controls, owing to the main effect of the causal variant, but that also differs between exposed and unexposed subjects, owing to the causal variant-exposure association. It is important to note that the spurious interaction cannot be removed by transformation of variables, as can be done in other cases,4 but is a direct result of measurement error of the causal variant. It does not depend on the mechanism of gene-environment dependence, of which Figures 1, 2, and 3 show a few examples, but arises from simple algebra of the statistical model.