Abstract. Although one-layer dynamic models of the Earth's mantle have successfully explained the geoid, they generate a surface dynamic topography that seems too large relative to geological observations. In this study, we hypothesize the possibility of partial advection of mantle equidensity surfaces by vertical motion induced by "driving" loads. These large-scale"flow-dependent" loads would greatly reduce the dynamic topography amplitude, while preserving a good fit to the observed geoid. Various physical processes related to nonequilibrium phase changes or to the existence of chemical heterogeneity in the mantle could justify a partial advection of the mean density. In this paper, we simply consider the flow-dependent loads as proportional to the vertical flow velocity. Two density mantle models a, re considered, one from subduction reconstruction [Ricard et al., 1993] and one from seismic tomography [Li and Romanowicz, 1995]. We show that a very moderate entrainment (a few kilometers) of the equidensity surfaces in the transition zone is sufficient to reduce dynamic topography amplitude by a factor of 2 or 3. The seismic velocity signal associated with this entrainment would be hidden by the signal of thermal origin. Using this new hypothesis, we compute sea level changes associated with epeirogeny for the Cretaceous, Paleocene, and Oligocene periods. The amplitude and phase of these changes are in fairly good agreement with geological hypsometric curves. Our results suggest that not only the thermodynamics, but also the kinetics of mineralogical phase changes in the transition zone are of crucial importance.