This chapter examines misspecification of conventional parametric human cap
ital models of earnings and presents nonparametric estimates of the earning
s function for both men and women. The chi-square tests reject the parametr
ic functional forms widely used in empirical analysis. Parametric models gi
ve very poor approximation to the true earnings function and their estimate
s are quite sensitive to the specification of parametric functional form. P
respecifying no functional forms, nonparametric models provide significantl
y better fit than the parametric models, especially for women's earnings fu
nction. Simple averages of earnings by age and education cells also provide
a relatively poor model of earnings, since the averages are subject to sig
nificant random fluctuations and have very high variances. Nonparametric re
gression models improve on the within-cell averages by smoothing across cel
ls.