Biological indicators are increasingly being used as integrative measures of ecosystem health in streams, particularly indicators using macroinvertebrate assemblage composition. Several indicators of this type have been advocated, including biotic indices based on taxa sensitivities, richness indices and ratios of observed to expected taxa from models predicting assemblage composition in streams with little human impact (O/E scores). The present study aimed to compare the sensitivity of indicators of each of these types (all used for legislated objectives for stream protection in Victoria, Australia) to a gradient of urban disturbance in 16 streams in a small area in eastern Melbourne. The biotic index SIGNAL and number of Ephemeroptera, Plecoptera or Trichoptera families were the most sensitive indicators, whereas total number of families and O/E scores from Australian river assessment system (AUSRIVAS) models were least sensitive. Differences in sensitivity were not the result of sampling or taxonomic inadequacies. AUSRIVAS and similar models might be improved by using only predictor variables that are not affected by human impacts and by sounder approaches to model selection. Insensitivities of indicators and misclassification of sites by the Victorian objectives show that assessment of indicators against disturbance gradients is critical for setting management objectives based on biological indicators.