The principles and practice of generalisation of cartographic elements in an automated system should not be assumed to be the same as those in manual cartography. Given the present lack of knowledge on what constitutes efficiently perceived graphics, generalisation procedures should now be re-designed as economically as possible in machine terms. Certain types of generalisation are, however, much more easily and usefully achieved than others by computer. These types are described in this text. The relative costs of generalisation and digitising of data at different input scales are also discussed.