This article investigates whether the returns to education vary with the level of cognitive ability. Unlike much of the literature, this article finds that the return to schooling is lower for those with higher cognitive ability indicating that education can act as a substitute for observed ability. Using quantile regressions we also find that, again unlike most of the literature, returns are higher at lower quintiles of the conditional earnings distribution. This suggests that education is also a substitute for unobserved ability. The policy implications are that increasing education in general and particularly for those with lower ability should reduce income inequality.