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 but smaller (and more precise) than the estimates typically reported by studies that use IV. Our results indicate that the biases generated by unobserved ability and measurement error tend to cancel each other out as suggested by Griliches (1977). We also present Monte Carlo evidence to show that the finite sample bias our estimator is small.