The sensitivity of eleven metrics using macroinvertebrate assemblages were evaluated in an environmental gradient in a tropical river in south-east Brazil. Eight sites were sampled in an altitudinal range of 160−650 m.a.s.l. during 1999 (April and August) and 2000 (February). Four substrates were sampled at each site: riffle litter, pool litter, stony substrates and sediment. Correspondence Analysis indicated that assemblages were primarily more influenced by physical changes (like deforestation and erosion processes) than the water chemistry. The sensitivity of each metric was evaluated through the application of box-and-whisker plot method by its power to assess impairment (metrics should be able to discriminate reference sites from impaired sites) and natural variability (metrics should not discriminate two reference sites). Metrics that failed in at least one of the above premises were not considered as sensitive. In this study, the most sensitive metrics were Shannon index, BMWP-ASPT, %_EPT, and relative abundance of EPT to Chironomidae.