We study the impact, efficiency and latent preferences of bank supervision with a structural model estimated on a newdataset ofwork hours of Federal Reserve supervisors. Using instrumental variables, we find that supervision has an economically large effect in lowering bank distress. The estimated supervisory cost function displays large economies of scale with respect to a bank’s size. Estimated supervisory preferences weight larger banks more than proportionally, consistent with macro-prudential objectives. We document a post-2008 reallocation of supervisory resources to large banks. We estimate that this reallocation decreased risk at large banks less than it increased risk at small banks, indicating a shift in preferences. We show evidence of frictions that prevent an efficient allocation of resources both within and across Federal Reserve districts. Model counterfactuals quantify the benefits of reducing these frictions, especially for the riskiest banks.