Estimates of the effect of education on GDP (the social return) have been hard to reconcile with micro evidence on the private return to schooling. We present a simple explanation combining two ideas: imperfect substitution and endogenous skill-biased technological progress and use cross-country panel data on inequality and GDP to test these ideas. A one-year increase in the level of education reduces the private return by 2 percentage points, consistent with Katz-Murphy's (1992) elasticity of substitution. We find no evidence for reversal of this initial effect as in Acemoglu (2002). In the short run, the social return equals the private return.