Numerous studies report gene-environment interactions, suggesting that specific alleles have different effects on social outcomes depending on environment. In all these studies, however, environmental conditions are potentially endogenous to unmeasured genetic characteristics. That is, it could be that the observed interaction effects actually reflect underlying genetic tendencies that lead individuals into certain environments. What is critical to move this literature forward is random environmental variation that we know is not correlated with innate characteristics of subjects. We exploit a natural experiment that randomizes a particular stressor--birth weight discordance within twin pairs--to address this challenge and ask: Do random differences in early environment (prenatal nutrition) moderate genetic effects on depression, delinquency, or GPA? Using Add Health data, the only consistently significant allele-birth weight interaction we reveal works in the opposite direction of Caspi et al.'s classic finding regarding the interaction of maltreatment with genetic variation in the serotonin transporter promoter. Less robust interactions found for "DRD2" and "MAOA" are consistent with this pattern that reverses prior findings. These results do not necessarily overturn existing research but support our methodological point that gene-environment research must address endogeneity. (Contains 6 tables, 1 figure and 9 notes.)