The decision to take more education is complex, and is influenced by individual ability, financial constraints, family background, preferences, etc. Such factors, normally unobserved by the researcher, introduce endogeneity and heterogeneity problems into estimating the returns to education. In this paper, these problems are addressed by estimating a comparative advantage model for schooling, in which the returns to education vary at different levels of education. The model requires that instruments must be specified at each level of education, and we suggest that different school reforms in Norway can serve as suitable instruments. In particular, we exploit the staged implementation of a major reform in the comprehensive school system in the 1960s. We find that the returns to education are strongly nonlinear. In particular, we find that the returns to upper secondary school and shorter programs at regional colleges, together with master?s programs at universities, have high returns as measured by wages. Also, we find that the average treatment effect is surprisingly high for medium-length educations (up to two years of college education). This means that increasing the general level of education, which was the intention of the comprehensive school reform of the 1960s and of other school reforms, has the potential to generate a high return in wages, although we do not consider the cost to society. We also find that there is a substantial difference between the average treatment effect and the effect of treatment on the treated for bachelor?s and master?s degrees at universities.