Approximately two-thirds of the tropical forest destroyed each year are caused by small-scale farmers that slash and burn the forest vegetation to plant crops and pastures. Characterisation and diagnosis of these land-use systems are the main theme of this 'Vegetation Preparatory Program' project based on three regions of the equatorial areas: Brazil, Peroo and Cameroon. To fulfil the requirements of the project, forest cover proportion was assessed from high (SPOT, landsat) and low (NOAA-AVHRR) resolution images. Classification of high resolution images in forest/crop land classes gave good results, but with some confusions between classes (open forest/dense forest for example). These confusions were reduced when the short wavelength infrared was included in the classification process. In comparison, classification of the low resolution images gave a poor estimation of the forest cover which is all the more over-estimated that the forest is fragmented. Forest cover proportion was then estimated from low resolution images, using spatial modelling and spectral mixture modelling, namely the TREES and COCKTAIL methods respectively. These two methods gave very close global values of deforestation at a regional scale, but the COCKTAIL method had the advantage to provide maps of forest cover, while TREES provided only a regional value.