Multi-temporal TM data, collected in 1987, has been processed in order to map twenty land-use categories over an area of 32 000 km2. Geometrical correction is accomplished by applying digital correlation techniques between TM images and a Ground Control Points (GCPs) data base. The data base, consisting of over 1000 GCPs, is composed of 15 by 15 pixels sub areas centred at UTM coordinates of all the control points used in previous studies. Principal components derived from multi-temporal data were subjected to unsupervised classification, and the spectral classes obtained were labelled using a semiautomatic process. The overall classification accuracy obtained is 85.6 per cent.