Identical twins long have been used to control for “ability” in efforts to obtain unbiased estimates of the earnings impact of schooling and of biases in estimates that do not control for earnings endowments. This study (1) presents new estimates of schooling returns and of “ability” bias using a new twins sample, (2) develops and applies a test of the significance of that bias, and (3) demonstrates that there may be “ability” bias even if the genetically-endowed component of ability does not affect schooling decisions directly as long as this component of ability is correlated with other family characteristics such as income that do affect schooling and that it is not possible to identify separately these individual components of “ability” bias. The basic empirical result is that, net of measurement error, upward “ability” bias is statistically significant in OLS estimates, causing an overestimate of the schooling impact of 12%.