This article sets out to demonstrate how boosting can serve as a supervised classification method, and to compare its results with those of conventional methods. The comparison begins with a theoretical example in which several criteria are varied: number of pixels per class, overlapping (or not) of radiometric values between classes, with and without spatial structuring of classes within the geographical space. The results are then compared with a real case study of land cover based on a multispectral SPOT image of the Sousson catchment area (South of France). It is seen that 1) maximum likelihood give better results than boosting when the radiometric values for each class are clearly separated. This advantage is lost as the number of pixels per class increases; 2) boosting is systematically better than maximum likelihood in the event of overlapping radiometric variable classes, whether or not there is a spatial structure.