This paper studies the puzzling lack of correlation between income and schooling in macro regressions. It is argued that the root of the puzzle is threefold. First, there is a problem of a proper definition of the way in which years of schooling should enter into a production function. Second, collinearity between physical and human capital stocks seriously undermines the ability of educational indicators to display any significance in growth regressions. Third, failure to cope with measurement error and endogeneity produces biased estimates. After dealing with these problems, the neoclassical approach to human capital is strongly supported by the data ...