We apply a new estimator to the measurement of the economic returns to education. We control for endogenous education, unobserved ability and measurement error using only the natural heteroscedasticty of wages and education attainment. Our prefered estimate, 6.07%, is closer to the OLS estimate...