The usefulness of biological indicators of stream condition depends on their sensitivity to human disturbance, and consistency of response across physiographic gradients. The Victorian State Environment Protection Policy legislates objectives for the biotic index SIGNAL, and the observed:expected (OE) index AUSRIVAS. AUSRIVAS is a poor predictor of human impact in the eastern Melbourne region, and while SIGNAL score is a sensitive index, the maximum SIGNAL score observed in the absence of human impacts varies across the physiographically diverse Melbourne region. Models predicting distributions of 60 macroinvertebrate families in relation to climatic, physiographic and human-impact variables were used to predict family occurrences in each reach of the region with human impacts set to zero. Using these predictions as the expected assemblage, I sought the OE index that a) best correlated with forest and urban land cover, b) showed the highest ratio of between-to-within-site variance in long-monitored sites of contrasting human impact, and c) was invariant across the region for predicted assemblages in the absence of human impacts. The best index weighted each observed and expected family at each site by its predicted probability of occurrence and by the shape and direction of its response to forest and urban impacts. This index, combining elements of OE and biotic indices, was as strongly correlated with human impacts as SIGNAL, but was less variable within sites, and had a consistent range of values across the region. Use of individual taxon distribution models to derive integrative indices of stream condition shows great promise.