A method is presented to derive surface albedo from METEOSAT radiances with a special application to Africa for January and June 1983 at 1130 UT. The method is based on radiative transfer simulations followed by a multivariate analysis. A linear relationship was derived between filtered (0.4-1.1 μm) surface albedo, METEOSAT radiance, optical depth at 0.55 μm, water vapor content and anisotropic conversion factor. After consideration of very different reflecting surfaces, anisotropic behavior of surface reflection is found to be dominant compared to atmospheric effects if extreme situations like dust outbreaks can be recognized and excluded. Multivariate analysis also revealed a linear relationship between filtered and unfiltered (0.3-2.6 μm) surface albedo. In this case atmospheric influence and anisotropy can be neglected. Errors of estimate, resulting from the applied regression model, are 0.02--0.04 if additional information about atmospheric state and anisotropy are available, for Sun zenith angles up to about 67¿ and for most of the viewing geometries. Application of the regression model to January and June 1983 at 1130 UT reveals that the surface albedo for Africa can be derived for low Sun zenith angles with reasonable accuracy from satellite radiances alone. The error of estimate is in the range of 0.02--0.03. Results derived with the proposed method are somewhat lower than satellite-derived surface albedos of the West African Sahel available from the literature. Comparison with in-situ data measured over six different surfaces in the Sahara shows good agreement in cases of large, homogeneous surfaces where point measurements can be assumed to represent larger areas. ¿American Geophysical Union 1991