We study optimal tax and educational policies in a dynamic private information economy, in which ex-ante heterogeneous individuals make an educational investment early in their life and face a stochastic wage distribution. We characterize labor and education wedges in this setting analytically and numerically, using a calibrated example. We present ways to implement the optimum. In one implementation there is a common labor income tax schedule, and a repayment schedule for government loans given out to agents during education. These repayment plans are contingent on loan size and income and capture the history dependence of the labor wedges. Applying the model to US-data and a binary education decision (graduating from college or not) we characterize optimal labor wedges for individuals without college degree and with college degree. The labor wedge of college graduates as a function of income lies first strictly above their counterparts from high-school, but this reverses at higher incomes. The loan repayment schedule is hump-shaped in income for college graduates.