Measures of short sale constraints and short selling activity strongly predict stock returns. This apparently exploitable predictability is difficult to explain. We partially resolve this puzzle by using measures of the stock borrowing costs paid by short-sellers. We show in portfolio sorts that the returns to short selling, net of stock borrowing costs, are much smaller than the gross returns to shorting or a typical long-short strategy. Option-implied borrowing fees, which reflect option market makers’ borrowing costs and the risks of changes in those costs, are on average only slightly higher than quoted borrowing fees. This finding indicates that the risk of changes in borrowing fee does not command a substantial risk premium. Option-implied borrowing fees predict future fees and stock returns, including returns net of quoted borrowing costs. The option-implied fee drives out other return predictors in panel regressions including option-based variables and other measures of short selling activity.