In the context of geographical database generalisation, this paper deals with a generic process for road network selection. It is based on the geographical context that is made explicit and on the characteristic structure preservation. It relies on literature work that is adapted and gathered. The first step is to detect significant structures and patterns of the road network such as roundabouts or highway interchanges. It allows to enrich the initial dataset with explicit geographic structures that were implicit in initial data. It helps both to explicit the geographical context and to preserve characteristic structures. Then, this enrichment is used as knowledge input for the following step that is the selection of roads in rural areas using graph theory techniques. After that, urban roads are selected by means of a block aggregation complex algorithm. Continuity between urban and rural areas is guaranteed by modelling continuity using strokes. Finally, the previously detected characteristic structures are typified to maintain their properties in the selected network. This automated process has been fully implemented on Clarity™ and tested on large datasets.