Note: This paper is dedicated to Aaron and Elizabeth Waters on the occasion of Dr. Waters' retirement.
Horizontal heat transfer, either by plates that cool as they move away from their source at a ridge axis or by convection currents, invalidates all temperature distributions calculated on the assumption of purely radial outward heat transfer. Temperature gradients in the upper mantle in oceanic regions are estimated on the assumptions that (1) the top of the low-velocity zone corresponds to the “wet” solidus of peridotite containing a very small amount of water, and (2) that the seismic discontinuity near 400-km depth corresponds to the olivine-β spinel inversion. The gradients are very low, suggesting that convection is the dominant mode of heat transfer throughout the upper mantle. Possible temperature profiles are accordingly drawn by joining a conduction solution in the lithosphere to an adiabatic (or near-adiabatic) curve below the lithosphere. The temperature profile at any point depends on the local age of the plate; it is impossible to devise a representative average oceanic geotherm that could be used, when superposed on a petrological phase diagram, to predict the depth of formation of magmas or to account for their compositional variations.