In this paper, I develop and implement a causal model of schooling and earnings, with heterogeneous treatment effects. The model specifies the connection between heteregeneous returns to schooling and family background variables. A control-function estimator is implemented using 2 types of exclusion restrictions. The analysis suggests that accounting for the heterogeneity in the returns to schooling due to differences in family background reduces the estimate of the average causal effect of education by 20%.