The south central Pacific Ocean is well covered with satellite-derived gravity data (ERS-1, GEOSAT, SEASAT and Topex/Poseidon) but sparsely covered with shipboard depth soundings and gravity measurements. We present an improved method of bathymetric prediction from gravity using both shipboard and satellite data. We employ a compensation model with two layers for the crust to attempt bathymetric prediction using a 2-D spectral approach. Parameters such as flexural wavelength and crustal density are the major constraints for the model. They are determined on a limited area by minimizing the difference between observed bathymetry ship profiles and the model predictions. Then, in regions where there are no ship data, these parameters are used to predict the seafloor topography from ship gravity data combined with satellite grid points. The prediction has an horizontal resolution limit of 8 km and the depth error is usually less than 300 m. This method is better constrained than the ones using satellite altimetry data only. Results on the Western Tuamotu archipelago and on Marquesas Islands, French Polynesia, are presented and discussed. ¿ American Geophysical Union 1996