This study presents evidence of heterogeneous returns to education over the wage distribution. The authors use instrumental variable quantile regression and data from the Swiss Labor Force Survey to identify the causal link between education and wages at different quantiles of the conditional distribution of wages. The results provide evidence that there is no unique causal effect of schooling and that for each individual the effect may deviate from those extensively documented by ordinary least squares or two-stage least squares. In particular, while ordinary quantile regression estimates increasing returns in the quantile index, once the endogeneity of schooling is taken into account the authors instead observe higher returns at lower quantiles of the wage distribution. Interpreting the quantile index as a measure of unobserved ability, the results suggest that higher-ability individuals have higher wages, but the slope of their wage-education profile is flatter than that for lower-ability individuals.