Keller [Keller, E.A., 1971. Areal sorting of bed-load material; the hypothesis of velocity reversal. Geological Society of America Bulletin 82, 753-756] hypothesized that at high flow, near-bed velocities in pools exceed velocities in riffles and create pool scour. Pools, however, typically have larger cross-sectional areas of flow at bankfull discharge. This condition raises an inconsistency with Keller's velocity reversal hypothesis and the one-dimensional continuity of mass equation. To address this problem, a model of pool maintenance and sediment sorting is proposed that relies on constriction of flow by recirculating eddies and flow divergence over the exit-slopes of pools. According to the model, a narrow zone of high velocity occurs in the center of pools, creating scour. Along the downstream end of pools, an uphill climb of particles up the pool exit-slope promotes sediment deposition. The model is tested with field and flume measurements of velocity, water-surface elevation, and size of bed sediments in recirculating-eddy influenced pools. Local reversals of the water-surface gradient were measured in the field and a velocity reversal was created in the flume. The reversals that were measured indicate higher gradients of the water surface over the upstream portions of pools and higher velocities in pools at high flow. The distribution of bed sediments collected in the field also support the proposed model of pool maintenance.