We study bank supervision by combining a theoretical model of asymmetric information and a novel dataset on work hours of Federal Reserve supervisors. We highlight the trade-offs between the benefits and costs of supervision and use the model to interpret the relation between supervisory efforts and bank characteristics observed in the data. More supervisory hours are spent on larger, more complex, and riskier banks. However, hours increase less than proportionally with bank size, suggesting technological scale economies in supervision. We provide evidence of constraints on supervisory resources, documenting reallocation of hours at times of stress and in the post-2008 period. Using variation implied by this resource reallocation, we find evidence that supervision lowers risk.