Our aim was to develop a simple method for testing gene-environment interaction in twin data ascertained through affected twins (probands), with known exposure status of both cotwins. To this end we derived formulae for two epidemiologic measures, as a function of prevalence of an exposure and genotype, and disease risk conditional on exposure and genotype: (1) relative risk of disease in exposed vs. unexposed cotwins, stratified by zygosity and proband exposure status (RRE), and (2) relative risk of disease in MZ vs. DZ cotwins, stratified by exposure status of proband and cotwin (RRZ). Then we investigated the behavior of these two measures under different assumptions about the relations between exposure and genotype in terms of their effect on disease risk. If an exposure has a different effect in the presence vs. absence of the genotype, RRE differs between MZ and DZ cotwins. Similarly, if a genotype has a different effect in exposed vs. unexposed persons, RRZ differs between exposed and unexposed cotwins. However, large differences in RRE between MZ and DZ cotwins, or large differences in RRZ between exposed and unexposed cotwins, do not occur except under very extreme conditions, such as a genotype that increases disease risk in exposed individuals, and has a protective effect in unexposed individuals. These results suggest that power to detect gene-environment interaction is limited with this approach. Conversely, if large differences in RRE or RRZ are observed, they are likely to reflect strong gene-environment interaction.