How do a regulator's decisions depend on the characteristics and strategies of its external clients? We develop a theory of approval regulation in which an uninformed regulator may veto the submission of a better-informed firm. The firm can perform publicly observable experiments to generate product information prior to submission. We find that when experimentation is short, Type I errors (approving bad products) are more likely for products submitted by firms with lower experimentation costs (larger firms), while Type II errors (rejecting good products) should be concentrated among smaller firms. These comparative statics are reversed when experimentation is long. We perform a statistical analysis on FDA approvals of new pharmaceutical products using two different measures of Type I error. We find consistent support for the counterintuitive hypothesis that, under particular conditions, errors are decreasing in the size of the firm submitting the product.