We consider the identification of the average treatment effect in models with continuous endogenous variables whose impact is heterogeneous. We derive a testable restriction that allows us to assess the degree of unobserved heterogeneity. Our analysis uses assumptions relating to the Local...