Twins-based estimates of the return to schooling feature prominently in the labor economics literature. The validity of such estimates hinges critically on the assumption that within-pair variation in schooling is explained by factors which are unrelated to wage earning ability. This paper develops a framework for testing this assumption, and finds, using a unique dataset of monozygotic twins, strong evidence against it. Di¤erences in adolescent IQ test scores predict within-pair variation in educational attainment, and including IQ in the wage equation causes within-pair point estimates for the returns to schooling to decline significantly. Our results thus cast doubt on the validity of estimates derived from the co-twin literature.