This paper examines the causal effect of parents' education on children's earnings using a model of endogenous schooling and earnings. The model suggests that parents' education is positively related to children's earnings, but its relationship with children's education is ambiguous. Identification is achieved by comparing the earnings of children with the same length of schooling, whose parents have different lengths of schooling. The model also features heterogeneous preferences for schooling and is estimated using HRS data. The empirically observed positive OLS coefficient obtained by regressing children's schooling on parents' schooling is mainly accounted for by the correlation between parents' schooling and children's unobserved preferences for schooling. This is countered by a negative, structural relationship between parents' and children's schooling choices, resulting in an IV coefficient close to zero when exogenously increasing parents' schooling. Nonetheless, an exogenous one-year increase in parents' schooling increases children's lifetime earnings by 1.2 percent on average. The key result from the model is that an individual’s schooling is a function of his learning ability and his parent’s human capital, while earnings is a function of his learning ability and schooling. This allows us to separately identify parental spillovers from selection by jointly using information on children’s schooling and earnings, in addition to parental background. The model suggests that the observed positive intergenerational schooling relationships indicate that selection effects are stronger than the level effect. It also explains why IV estimates of intergenerational schooling find no effect. The negative level effect is key, a feature we do not artificially assume, but inherent in most BP life-cycle models of human capital accumulation. Reduced-form evidence from the HRS data reveals that children with the same years of schooling have parallel earning profiles with constant gaps, which are increasing in their moms' years of schooling. The gaps indicate that moms affect how much human capital a child accumulates before labor market entry, compared to another child with the same years of schooling. The fact that these earnings profiles are parallel suggests that spillovers no longer operate once children enter the labor market. The size of the gaps is similar whether we look at children who do or don't finish high school, or children who do or don't enroll in college. All such evidence is consistent with our model assumptions. We estimate an extended model by indirect inference. We use moms' schooling as a proxy for parents' human capital. We allow children's human capital production technology to differ depending on whether he is in school or is working, and also allow for heterogeneous, non-pecuniary preferences for schooling. Including preference heterogeneity not only helps replicate key moments in the data but also prevents the overestimation of the effects of parents' human capital and children's learning abilities on children's schooling and earnings outcomes. Preferences are identified by an exclusion restriction: abilities affect both schooling and earnings outcomes, whereas preferences exclusively affect schooling choices. Empirically, the reduced-form effect of moms' schooling on children's earnings is tooThis paper examines the causal effect of parents' education on children's earnings using a model of endogenous schooling and earnings. The model suggests that parents' education is positively related to children's earnings, but its relationship with children's education is ambiguous. Identification is achieved by comparing the earnings of children with the same length of schooling, whose parents have different lengths of schooling. The model also features heterogeneous preferences for schooling and is estimated using HRS data. The empirically observed positive OLS coefficient obtained by regressing children's schooling on parents' schooling is mainly accounted for by the correlation between parents' schooling and children's unobserved preferences for schooling. This is countered by a negative, structural relationship between parents' and children's schooling choices, resulting in an IV coefficient close to zero when exogenously increasing parents' schooling. Nonetheless, an exogenous one-year increase in parents' schooling increases children's lifetime earnings by 1.2 percent on average. The key result from the model is that an individual’s schooling is a function of his learning ability and his parent’s human capital, while earnings is a function of his learning ability and schooling. This allows us to separately identify parental spillovers from selection by jointly using information on children’s schooling and earnings, in addition to parental background. The model suggests that the observed positive intergenerational schooling relationships indicate that selection effects are stronger than the level effect. It also explains why IV estimates of intergenerational schooling find no effect. The negative level effect is key, a feature we do not artificially assume, but inherent in most BP life-cycle models of human capital accumulation. Reduced-form evidence from the HRS data reveals that children with the same years of schooling have parallel earning profiles with constant gaps, which are increasing in their moms' years of schooling. The gaps indicate that moms affect how much human capital a child accumulates before labor market entry, compared to another child with the same years of schooling. The fact that these earnings profiles are parallel suggests that spillovers no longer operate once children enter the labor market. The size of the gaps is similar whether we look at children who do or don't finish high school, or children who do or don't enroll in college. All such evidence is consistent with our model assumptions. We estimate an extended model by indirect inference. We use moms' schooling as a proxy for parents' human capital. We allow children's human capital production technology to differ depending on whether he is in school or is working, and also allow for heterogeneous, non-pecuniary preferences for schooling. Including preference heterogeneity not only helps replicate key moments in the data but also prevents the overestimation of the effects of parents' human capital and children's learning abilities on children's schooling and earnings outcomes. Preferences are identified by an exclusion restriction: abilities affect both schooling and earnings outcomes, whereas preferences exclusively affect schooling choices. Empirically, the reduced-form effect of moms' schooling on children's earnings is too