We estimate a Dynamic Programming model of the decision between continuing schooling or entering the labor market using a panel from the National Longitudinal Survey (NLSY). The model, set in an expected utility framework (with a power utility function), fits data on both schooling attainments and wage very well. We find a degree of relative risk aversion much smaller than usually found in the finance literature (around 0.6) and a subjective annual discount rate between 4\% and 5\%. Various simulations indicate that schooling attainments are elastic with respect to the return to college education and, to a lesser extent, with respect to measured ability (Armed Forces Qualification Test scores) but inelastic with respect to household characteristics (especially household income). The ``true’’ intergenerational correlation between schooling attainments and parents’ education (after conditioning on observed and unobserved ability) is found to be quite low. Finally, our estimates of the return to schooling indicate strong returns to high school graduation and a clear evidence of a positive correlation between the utility of attending school and unobserved labor market ability (ability bias). Estimates of the return to schooling which take into account the ability bias are found to be between 25% and 30% smaller than those obtained ignoring it.