We estimate the time-averaged velocity field in the North Atlantic from observations of density, wind stress and bottom topography. The flow is assumed geostrophic, with prescribed Ekman pumping at the surface, and no normal component at the bottom. These data and dynamics determine velocity to within an arbitrary function of (Coriolis parameter)/(ocean depth), which we call the “dynamical free mode.” The free mode is selected to minimize mixing of potential density at mid-depth. This tracer-conservation criterion serves as a relatively weak constraint on the calculation. Estimates of vertical velocity are particularly sensitive to variations in the free mode and to errors in density. In contrast, horizontal velocities are relatively robust. Below the thermocline, we predict a strong O (1 cm/sec) westward flow across the entire North Atlantic, in a narrow range of latitude between 25N and 32N. This feature supports the qualitative (and controversial) conjecture by Wüst (1935) of flow along the “Mediterranean Salt Tongue.” Along continental margins and at the Mid-Atlantic Ridge, predicted bottom velocity points along isobaths, with shallow water to the right. These flows agree with many long-term current measurements and with notions of the circulation based on tracer distributions. The results conflict with previous oceanographic-inverse models, which predict mid-depth flows an order of magnitude smaller and often in opposite directions. These discrepancies may be attributable to our relatively strong enforcement of the bottom boundary condition. This involves the plausible, although tenuous, assertion that the flow “feels” only the large-scale features of the bottom topography. Our objective is to investigate the consequences of using this hypothesis to estimate the North Atlantic circulation.