This paper surveys various econometric issues that arise in estimating a relation between the logarithm of earnings, schooling, and other variables and focuses on the problem of "ability" as a left-out variable and the various solutions to it. It points out that in optimizing models the "ability bias" need not be positive and shows, using recent analyses of NLS data, that when schooling is treated symmetrically, allowing it too to be subject to errors of measurement and correlated to the disturbance in the earnings function, the usual conclusion of a significantly positive "ability bias" in the estimated schooling coefficients is not only not supported but possibly even reversed.