The very long wavelength (i.e. in the degree range ℓ=1--8) dynamic topography at the Earth's surface, estimated by isostatic reduction of the observed topography, is shown to be in good agreement with the topography, is shown to be in good agreement with the topography predicted by a model of viscous flow in the mantle. This flow model was previously derived by independent consideration of the observed nonhydrostatic geoid and is based upon a recent model of large-scale seismic shear velocity heterogeneity in the mantle. Our estimate (and prediction) of the long wavelength dynamic surface topography reveals significant depressions (≈2--3 km) of the continental shields relative to their hydrostatic reference positions. ¿ American Geophysical Union 1993