Automatic interpretation and digitization of scanned maps is an active field of investigation, since the data included in maps are needed for the constitution of geographical information systems or to improve the analysis of far more complex documents like satellite images and radar data. The authors deal with automatic interpretation of aerial images, for the constitution of 1:25,000 scale maps. One comes up against a diversity of contexts to handle and a complexity of information to extract. External data from an existing map is a promising aid to reach exhaustive and operational results. This involves an additional step of map digitization. But the advantages are worth the effort as the map contains an available complete analysis of the scene, performed during its conception. A selection is then carried out, keeping useful information, smoothing shapes during the generalization step and rejecting details (bushes, hedges, small structures, non-static phenomena). It provides global comprehension of the area in the aerial image, involving network organization, spatial relationships and an exhaustive inventory of meaningful geographical objects.