Land cover changes are measured at increasingly broader spatial scales. Yet understanding and modelling change processes with a satisfactory accuracy require fine scale observations. The objective of this study is to design and test a methodology to detect tropical deforestation 'hot spots' at broad spatial scales. This methodology is designed to allow concentration of the data collection and modelling of change processes only on the areas characterized by a high rate of change. The procedure is based on a hierarchical set of decision rules with selection criteria being first measured on an exhaustive basis at a global scale and then only for the areas retained in the first sorting, with increasingly selective constraints. The first set of criteria, i.e., proportions in key land cover, landscape fragmentation, and fire activities, were derived from subcontinental scale remote sensing data. Socio-economic variables were also measured at that scale. These different variables were combined over West Africa and the northern boundary of the Central African evergreen forest to identify potential tropical deforestation fronts. Different models were used to generate maps of deforestation hot spots. These were validated with data from the literature.