Analyses of the long-wavelength geoid with seismic tomographic models have been providing for a long time important estimates of mantle viscosity. These estimates have nearly been derived under the assumption of whole mantle flow. It has been commonly held that a fully impermeable boundary at 660 km depth is incompatible with the long-wavelength gravity signal. On the other hand, models with whole mantle circulation, which can explain a large portion of the geoid signal, usually produce excessive amplitudes of the dynamical topography, especially for long wavelengths. Using recent tomographic models together with genetic algorithm we have successfully demonstrated that the layered convection model can also produce a reasonable fit to the geoid, which is comparable in quality with that obtained for the whole mantle model. The layered model can simultaneously yield realistic amplitudes of the dynamical topographies of the surface and the 660-km discontinuity.¿ 1997 American Geophysical Union