Mapping alpine vegetation at a meso-scale (catchment level) using remote sensing presents difficulties due to a patchy distribution and heteregeneous spectral appearance of the plant cover. We discuss issues of generalization and accuracy assessment in this case study when using a digital CIR air photo for an automatic classification of the dominant plant communities. About 10 plant communities could be classified with acceptable accuracy where the chosen classification scheme determined the final outcome. If a high resolution pixel mosaic is generalized to units that match the positional accuracy of simple GPS this generalization may also influence the information content of the image