Growing pressures from development activities, on aquatic environments in South American countries, have created an urgent need for tools to measure the quality of aquatic systems. The index of biotic integrity, based on fish assemblages, elaborated in North America, constitutes a basis for the development of these tools. However, its direct application is problematic in regions having a rich and diverse ichthyofauna and where the knowledge of fish species ecology is incomplete. The response of taxonomic or trophic groups is often unknown and the selection of fish assemblage metrics cannot be based on previous knowledge, as in intensely studied regions. This work proposes a method, similar to the approach recently pursued in Europe, based on comparing the variability of fish assemblages in reference situations with their variability influenced by human-induced perturbations. The method was developed from 53 samples of 27 reaches in 7 hydrographic basins of French Guiana. For each of 28 fish assemblage descriptors, stepwise multiple linear regressions with 28 habitat variables were carried out. The residuals of the models obtained were used as candidate metrics independent of natural environmental factors. Nine metrics showing significant differences between reference and disturbed samples were selected to constitute the index. The index was validated by analysing a temporal data series obtained from a reach disturbed by dam construction.