This paper proposes a nonparametric method of estimating marginal treatment effects in heterogeneous populations. Building upon an insight of Heckman and Vytlacil, the conventional treatment effects model with heterogeneous effects is shown to imply that outcomes are a nonlinear function of participation probabilities. The degree of this nonlinearity, and hence the shape of the marginal response curve, can be estimated with series methods such as power series or splines. An illustration is provided for the returns to higher education in the U.K, indicating that marginal returns to higher education fall as the proportion of the population with higher education rises, thus providing evidence of heterogeneity in returns.