In most studies, genotypes are measured not for the variants that directly affect disease but for markers in linkage disequilibrium (LD) with the causal variants. This is especially true in GWASs and other large-scale discovery studies that aim to map novel disease variants. This creates a misclassification problem in that the true causal variants have been measured with error. In contrast to common measurement error models, a marker is not an unbiased measurement of a causal variant (because the genotype frequencies of the marker and causal variant may differ), and the misclassification probabilities are unknown by design. General methods to adjust for measurement error6 are not applicable, and so we must accept possible bias in the estimation of interaction effects.