This paper is a discussion of the short-wavelength (25--100 km) undulations of the geoid recovered from Seasat altimeter data in the South Atlantic. These undulations are known to be highly correlated to seafloor topography, and they considerably improve our knowledge about global bathymetry of the South Atlantic, which remains poorly surveyed. The orientation of fracture zones (FZ) with respect to the Seasat profiles ensures an efficient recovery of the FZ pattern in the younger (<65 Ma) parts of the South Atlantic between 20¿N and 50¿S. We used 22 well-marked FZs reguarly distributed between 12¿N and 47¿S to compute Eulerian poles describing the finite relative movement between Africa and South America. From the present back to 35 Ma. FZ trends south of the equator are well accounted for with a single rotation, and standard statistical tests do not indicate a need for distinct and successive rotations. However, these tests indicate an improvement of the fit when distinct poles are adjusted to separate subsets of our 22 FZs (either FZs north and south of the equator, or FZs north and south of Tristan da Cunha). This points out the possibility of the existence of either ''corridors'' in which slight relative movements occur, or diffuse elastic deformation over the entire South Atlantic, although this is really a second-order effect. We also present and discuss a map of the roughness of the geoid (for the 25- to 110-km wave band). Strong disagreements between the roughness map and the GEBCO maps exist in regions of sparse ship track coverage; for example, the Rio Grande Rise has an associated roughness feature which appears elongated and more symmetric with respect to the Walvis Ridge than what is displayed on the GEBCO maps. The roughness map also clearly documents a long-wavelength (~400 km) segmentation of the Mid-Atlantic Ridge between the equator and 35¿S. ¿ American Geophysical Union 1989