We examine evidence on omitted-ability bias in estimates of the economic return to schooling, using proxies for unobserved ability. We consider measurement error in these ability proxies and the potential endogeneity of both experience and schooling, and examine wages at labor market entry and later. Including ability proxies reduces the estimate of the return to schooling, and instrumenting for these proxies reduces the estimated return still further. Instrumenting for schooling leads to considerably higher estimates of the return to schooling, although only for wages at labor market entry. This estimated return generally reverts to being near (although still above) the OLS estimate if we allow experience to be endogenous. In contrast, for observations at least a few years after labor market entry, the evidence indicates that OLS estimates of the return to schooling that ignore omitted ability are, if anything, biased upward rather than downward.