Roads are essential component of topographic maps and spatial databases. The challenge in automated generalisation of road networks is to derive a connected network while maintaining the structure for the intended target scale and to achieve this with minimum user intervention. A lot of methods to select, displace and simplify roads have been presented; the focus here is on the generalisation of networks using visual perception techniques. This paper presents a framework based on visual perception that uses minimum attributes for generalisation of both ‘rural’ and ‘urban’ roads over large scale change. The system incorporated graph theoretic techniques to explicitly model the topology of the network as it was generalized. The model uses a fine scale map (1:1250 or 1:2500) as input and generates small scale (1:250,000) maps directly from it without creating intermediate small scale maps. The results compared favorably with paper maps (Ordnance Surveys Stratgie dataset (1:250,000)).