A subduction shear zone can be modeled as a long narrow channel, with the flow of subducted metasedimentary rocks in the channel driven by two sets of forces: the downward shearing force exerted by the subducting slab and the gradient in the hydraulic potential, which combines the effect of both pressure and buoyancy. If the channel walls are effectively rigid, very slight narrowing or broadening of the channel (convergence angles <1°) can result in very dramatic changes in the (nonlithostatic) pressure distribution along the channel. The geometry of the subducting plate, which is forced to bend under the overriding plate, suggests that the channel should initially narrow downward and then gradually broaden. A model assuming this geometry, with initial channel width 1500 m, minimum width ∼600 m and width at 100 km depth of again ∼1500 m, a maximum viscosity of 1019 Pa s, and a convergence rate of 8 cm/yr reaches pressures >2 GPa in the channel at only 40 km depth. The model is consistent with a horizontal balance of forces across the plates and with a reasonable value for the thickness of subducted sediment (∼650 m). The practical limit for overpressures attainable in subduction zones is determined by the strength and permeability of the channel walls. At 40 km depth the channel is effectively confined on both sides by cold lithospheric mantle, which should be strong enough to support a significant tectonic overpressure. Episodic failure of the upper plate to produce great earthquakes at 30–40 km focal depth could vent overpressured fluid from the channel, allowing a cyclical buildup and release of both rock and fluid pressure. Topography on the subducting plate (e.g., seamounts and thinned continental crust) may lead to an anvil-like jamming of the channel and local high overpressures. Tectonic erosion by topography on the lower plate of slivers from overlying continental crust and the compression of these slivers between the topography and the narrowing channel walls could produce high overpressures in continental rocks. A decrease in the convergence rate or cessation of subduction, with a consequent general warming within the channel and associated viscosity decrease, promotes exhumation by buoyant reverse flow. The most rapid reverse flow occurs in the region of previously greatest overpressure. Since the exhumation distance is shorter than for a simple lithostatic pressure distribution and any increase in temperature is coupled with a strong increase in the rate of exhumation, preservation of high-pressure assemblages at the surface in fossil subduction zones is promoted for such a model.