Description: Large-scale topography may be due to several causes, including (1) variations in crustal thickness and density structure, (2) oceanic lithosphere age differences, (3) subcrustal density variations in the continental lithosphere and (4) convective flow in the mantle beneath the lithosphere. The last contribution in particular may change with time and be responsible for continental inundations; distinguishing between these contributions is therefore important for linking Earth's history to its observed geological record. As a step towards this goal, this paper aims at such distinction for the present-day topography: the approach taken is deriving a ‘model’ topography due to contributions (3) and (4), along with a model geoid, using a geodynamic mantle flow model. Both lithosphere thickness and density anomalies beneath the lithosphere are inferred from seismic tomography. Density anomalies within the continental lithosphere are uncertain, because they are probably due to variations in composition and temperature, making a simple scaling from seismic to density anomalies inappropriate. Therefore, we test a number of different assumptions regarding these. As a reality check, model topography is compared, in terms of both correlation and amplitude ratio, to ‘residual’ topography, which follows from observed topography after subtracting contributions (1) and (2). The model geoid is compared to observations as well. Comparatively good agreement is found if there is either an excess density of 0.2 per cent in the lithosphere above 150 km depth, with anomalies below as inferred from tomography, or if the excess density is 0.4 per cent in the entire lithosphere. Further, a good fit is found for viscosity 10 20 Pa s in the asthenosphere, increasing to 10 23 Pa s in the lower mantle above D'. Results are quite dependent on which tomography models they are based on; for some recent ones, topography correlation is 0.6, many smaller scale features are matched, topography amplitude is less than 30 per cent too large, while geoid variance reduction exceeds 70 per cent—overall a considerable improvement compared to previous models. Correlation becomes less if smaller scale features (corresponding to spherical harmonic degrees 15 and higher), which are probably largely due to anomalies in the lithosphere, are also considered. Comparison of results with different viscosity structures, and a regional comparison of amplitude ratios, indicates that lateral viscosity variations can be quite strong, but only leading to moderate variations in model topography of a factor probably less than two.