Comparison between the number of taxa observed and the number expected in the absence of human impact is an easily understood and ecologically meaningful measure of biological integrity. This approach has been successfully applied to the assessment of the biological quality of flowing water sites using macroinvertebrates with the river invertebrate and classification system (RIVPACS) and its derivatives. In this paper, we develop a method similar to the RIVPACS predictive model approach to assess biological integrity at flowing-water sites using freshwater fish and decapod assemblages. We extend the RIVPACS approach by avoiding the biotic classification step and model each of the individual species separately. These assemblages were sampled at 118 least impacted (reference) sites in the Auckland region, New Zealand. Individual discriminant models based on the presence or absence of the 12 most common fish and decapod species were developed. Using the models, predictions were made using environmental measures at new sites to yield the probability of the capture of each of the 12 species, and these were combined to predict the assemblage expected at sites. The expected assemblage was compared to that observed using an observed over expected ratio (O/E). The models were evaluated using a number of internal tests including jackknifing, data partitioning, and the degree to which O/E values differed between reference sites and a set of sites perceived to be impaired by human impacts.