Land cover information plays an elementary role for regional water and land management, and is an essential variable for the assessment of ecosystem services and regional climate impact. This paper describes the generation of a regionally optimized land cover dataset for West Africa with a spatial resolution of 250 m, which is based on earth observation data from three optical and radar instruments. The choice of sensors is based on their individual strengths and weaknesses in assessing specific land surface types. Annual profiles of the optical Moderate Resolution Imaging Spectroradiometer (MODIS) are analyzed for the classification of vegetated classes including agriculture. The classification approach builds on random forest classification with learning data extracted from higher resolution land cover maps. Envisat Advanced Synthetic Aperture Radar (ASAR) Wide Swath (WS) time series are used, in combination with MODIS data, to delineate permanent and seasonal water bodies. Here, an approach integrating threshold classification and morphological operations is applied. Built-up areas of different densities are identified based on a seamless coverage of radar imagery collected by the satellites TanDEM-X and TerraSAR-X. The detection of settlements is based on an unsupervised classification scheme which exploits texture metrics and backscattering amplitudes of the fine resolution radar sensors. The accuracy assessment of the multi-sensor land cover map yields an overall accuracy of 80% at legend level 1 (9 classes) and 73% at the more detailed legend level 2 (14 classes). Comparisons with available wall-to-wall datasets of the region demonstrate the valuable information content of the presented West African land cover map.