In this paper, we investigate the potential of large-scale mosaics of Synthetic Aperture Radar (SAR) data for land cover mapping. The study is based on a wall-to-wall mosaic of double-polarization data (HH and HV) from the L-band sensor PALSAR, covering the whole African continent at a spatial resolution of about 100m. The GlobCover 2009 global land cover map is taken as a reference for the training of the classification algorithm. The joint use of the PALSAR mosaic, the GlobCover 2009 map and the Shuttle Radar Topography Mission (SRTM) Digital Elevation Model (DEM) allows retrieving backscatter signatures that take into account local topography (slope angle and slope orientation) for each of the 22 GlobCover classes. These signatures are then used to assign each pixel of the mosaic to one class. The 22 classes are finally combined to coarser classes: tree cover, shrub cover, herbaceous cover/bare soil, and water. Because of ecological and phenological variations over the continent, this land cover mapping scheme is actually applied to 5° by 5° tiles. The methodology has been developed so far on two tiles which illustrate different ecoregions. A visual assessment indicates that the classified maps are very satisfying, at least for the tree cover class, with a significantly improved spatial resolution compared to existing large-scale land cover products. Validation is ongoing before applying the method to the whole Africa in the future.