The basis of this theoretical paper is the assumption that the asthenosphere is ‘soft’ enough so that it does not transmit significant horizontal shear stress to the overlying lithosphere. It is pointed out that any thermal mantle convection is likely to have its origin in the cooling of the top layers of the oceanic mantle rather than being due to any heating that may arise farther down. Next, there is geological evidence for the existence of regions that exhibit widespread horizontal tension, e.g., in East Africa. Submarine ridges can justifiably be considered as regions of horizontal tensile stress. In our model, the choice of mechanisms that can generate such stresses is severely limited; the one found to be satisfactory qualitatively and quantitatively arises through the disappearance of pieces of lithosphere, which slide down along the Gutenberg-Benioff fault zones. The lithosphere spreads to fill the gaps created. It is shown that this motion is likely to represent the surficial part of a thermally driven conveetive circulation in the mantle. The nature of the ‘return flow’ farther down is discussed.