This paper explains the methodology and results of research on the growth of large forest fires, using information from the NOAA-AVHRR sensor. Numerous algorithms have been developed for discriminating pixels of active fires from the rest. These can be divided into three groups: a) those based on information from channel 3 (single channel); b) those using algorithms from several channels (multi-channel); c) those that, as well as using this information, compare the pixel with those around it (contextual). This research used the contextual algorithm of Flasse and Ceccato1 to discriminate between pixels corresponding to active fires and the rest, and to interpret the growth of the fire. Two fires were selected for the purpose: those in the Sierra de Cazulas (Granada) in August 1999, and in Valencia, in July 1994. The results provided useftil information on the growth of fire in general, the area, speed at which it spread and prevailing direction. This suggests that this method can be recommended for studies on fire behaviour, as well as in the provision of resources for fighting the fire at both the regional and national level.