Banks often charge implicitly for their services via interest spreads, instead of explicit fees. Much of bank output thus has to be estimated indirectly. In contrast to current statistical practice, dynamic optimizing models of banks argue that compensation for bearing systematic risk is not part of bank output. We apply these models and find that in the U.S. National Accounts between 1997 and 2007, bank output was overestimated by 21% and gross domestic product (GDP) by 0.3%. Compared with current methods, our new estimates imply more plausible estimates of the income share of capital and the return on fixed capital of the banking industry. (JEL E01, E44, O47)