To examine the topography and gravity anomalies due to mantle convection, we have carried out finite difference calculations of thermal convection in a fluid layer with a viscosity exponentially decreasing with temperature. Both surface topography and gravity anomalies are shown to be positive over regions of ascending flow and negative over regions of descending flow. These results differ significantly from those of McKenzie (1977) which for similar conditions predict negative topography and gravity anomalies above rising plumes. At large Rayleigh number, the amplitude of surface topography is found to depend on Rayleigh number to the seven-ninths power as predicted by boundary layer theory. These results are applied to test the hypothesis that the linear small-scale gravity undulations in the Central Pacific Ocean (Haxby and Weissel, 1983) are caused by convective rolls in a layer at the base of the lithosphere. For a convecting layer thickness one-half the observed gravity wavelength and with plausible values of flexural rigidity and heat flux, we show that convection can produce gravity anomalies of the observed magnitude with a layer viscosity comparable to that determined by post-glacial rebound. However, a large increase of viscosity with depth is required to confine convection to a thin layer. If these anomalies are actively maintained by convective stresses, one possibility is that layered convection may result from compositional stratification of the mantle.