This work describes spectral characteristics of fire scars in an area of intense deforestation in the Amazon tropical forest as recorded in the seven Landsat-5 TM spectral channels of an image in 1985. Ground covers were divided into four themes: natural forests, ‘capoeiras’, pastures in general and fire scars. Single-cell and maximum likelihood classification algorithms were used in the analysis. TM channel 4 was the best to identify fire scars, with digital counts differences in relation to other themes about two times higher than in other channels. Fire scars were also detected in channel 5, with the limitation that recent scars could be mistaken with water. Channels 3 and 7 also contained information about the scars. Results indicated that TM images can be used in automatic detection and assessment of biomass burning in tropical forests, contributing to the understanding of biogeochemical cycles related to the use of fire in these areas. Comparison of the results for channels 3, 4 and 5 in two more TM scenes in 1987 and 1989, showed that minor variations occurred in the spectral definition of the themes, mainly as a result of atmospheric pollution from biomass burning.