The behavior of interest rates under the U.S. National Banking System is puzzling because of the apparent presence of persistent and large unexploited arbitrage opportunities for note issuing banks. Previous attempts to explain interest rate behavior have relied on the cost or the inelasticity of note issue. These attempts are not entirely satisfactory. Here we propose a new rationale to solve the puzzle. Inelastic note issuance arises endogenously because the marginal cost of issuing notes is an increasing function of circulation. We build a spatial separation model where some fraction of agents must move each period. Banknotes can be carried between locations; deposits cannot. Taking the model to the data on national banks, we find it matches the movements in long-term interest rates well. It also predicts movements in deposit rates during panics. However, the model displays more inelasticity of notes issuance than is in the data.