This paper uses new administrative data with detailed borrower information and lengthy repayment histories from the Canada Student Loans Program (CSLP) to measure rates of return on undergraduate student loans. We document substantial heterogeneity in returns based on information available at the time loans were disbursed, including province of residence, field of study, and institution of attendance. Field of study is a particularly important determinant of rates of return, explaining 22% of the variation in predicted returns across borrowers. We explore the implications of this variation for CSLP cross-subsidization across borrowers and potential risk-based loan limits. Given the variation in ex ante predicted returns across borrowers, using all available information at the time of loan disbursement, we study the implications of potential cream-skimming of high-return borrowers by private lenders.