Weconstruct a set of human capital indexes for the states of theUnited States for each census year starting in 1940. To do so,we propose a new methodology for the construction of index numbersin panel data sets. Our method is based on an optimal approachby which we choose the ``best'' set of index numbers by minimizingthe expected estimation error subject to some research constraints.Some of the empirical findings are that the stock of human capitalin the United States grew twice as rapidly as the average yearsof schooling and that human capital inequality across stateswent up during the 1980s (while the dispersion of schooling actuallyfell). We conclude that using the average years of schoolingfor the empirical study of existing growth models may be misleading.