The Ordnance Survey has traditionally recorded the large-scale topography of Britain as Cartesian co-ordinate-based point, line and text label features within the tile-based Land-Line® Database. Under their Digital National Framework™ (DNF™) project, this data has been re-engineered into a topologically structured format known as OS MasterMap™ [Ordnance Survey]. This required the modelling of the area features enclosed by the line data as polygon objects. This new polygon-enriched data can be provided seamlessly for pre-defined areas and by theme. Each feature is assigned a unique Topographic Identifier (TOID™) number, allowing for the easy updating of a data holding, and the association of any topographic feature with external information. Each point object is classified with a particular feature code, such as post-box or bench-mark; likewise, a line feature could be labelled as a building outline or a public road edge. The feature-coding of polygons is the most difficult requirement of the DNF, as it requires the inferring of information that is not present in the Land-Line data. Properly classified area features greatly add to the intelligence of the resulting OS MasterMap data, allowing a myriad of valuable analyses to be carried out. The OS has accomplished high quality polygon classification semi-automatically, largely by examining the feature codes of the lines that bound each polygon. Using novel feature-coding techniques, the accuracy can be further improved. Work is continuing within our research group on the application of computer vision techniques to polygon classification through shape recognition [Keyes and Winstanley]. The results of several shape descriptors have been fused and good results have been achieved. By combining the results with other classification techniques, a more robust feature-coding tool can be developed. In this paper, the classification of polygons based on their context is described.