Tectonic features at the earth's surface can be used to test models for mantle return flow and to determine the geographic pattern of this flow. A model with shallow return and deep continental roots places the strongest constraints on the geographical pattern of return flow and predicts recognizable surface manifestations. Because of the progressive shrinkage of the Pacific (averaging 0.5 km2/yr over the last 180 m.y.) this model predicts upper mantle outflow through the three gaps in the chain of continents rimming the Pacific (Carribbean, Drake Passage, Australian-Antartic gap). In this model, upper mantle return flow streams originating at the western Pacific trenches and at the Java Trench meet south of Australia, filling in behind this rapidly northward-moving continent and provding an explanation for the negative bathymetric and gravity anomalies of the 'Australian-Antarctic-Discordance'. The long-continued tectonic movements toward the east that characterize the Caribbean and the eastenmost Scotia Sea may be produced by viscous coupling to the predicted Pacific outflow through the gaps, and the Caribbean floor slopes in the predicted direction. If mantle outflow does not pass through the gaps in the Pacific perimeter, it must pass beneath three seismic zones (Central America, Lesser Antiles, Scotia Sea); none of these seismic zones shows foci below 200 km. Mantle material flowing through the Caribbean and Drake Passage gaps would supply the Mid-Atlantic Ridge, while the Java Trench supplies the Indian Ocean ridges, so that deep-mantle upwellings need not be centered under spreading ridges and therefore are not required to move laterally to follow ridge migrations. The analysis up to this point suggests that upper mantle return flow is a response to the motion of the continents. The second part of the paper suggests driving mechanism for the plate tectonic process which may explain why the continents move. This hypothetical driving mechanism has four aspects: (1) the lower mantle convects, (2) this convection drives the continents by drag on their deep roots. (3) return flow in the upper mantle provides volumetric balance for the motion of the continental masses without passing under them, and (4) oceanic plates are effectively decoupled from the asthenosphere and are driven largely by slab pull. This mechanism accounts for the opening of the Atlantic, the ability of spreading ridges to stay on the midlines of oceans, and the penetration of India into Asia following their collision. Continental motions strongly imply lower mantle upwelling and divergence beneath the Atlantic and southeast Indian Oceans, and convergence and subsidence along the Tethyan belt. This picture disagrees with the concept, derived from hot spot studies, of an underforming, absolute reference framework at depth, but weakness in current hot spot theory would make a rejection of the present model on this ground premature. If the present model is generally correct, a fairly simple pattern of lower mantle convection, with as few as four cells, may explain much of the tectonic complexity of the earth.