We have increasingly used predictive models in the USA to improve characterization of site-specific reference conditions and interpretation of biological responses to environmental alteration. In general, we have found that (1) Random Forest models are more accurate and precise than other modeling techniques we have evaluated, (2) variation in taxa composition is well predicted from a few temperature and flow variables, and (3) model precision is likely constrained by regional variation in meta-community dynamics. In this talk I describe how we have applied some of these advances to examine how stream invertebrate biodiversity might respond to climate change. We predicted current and future stream temperature and flow regimes by linking climate model predictions to Random Forest models that predict stream temperature and hydrologic regime from climate and watershed attributes. We then linked stream temperature and flow predictions to a RIVPACS-type model to predict how site-specific probabilities of capture of 539 benthic invertebrate taxa would change at 1197 stream reaches in response to climate-induced alteration in stream temperature and flow regime. Both individual taxa and assemblages were predicted to vary markedly in their vulnerability to climate change. Differences in predicted taxa vulnerabilities were generally consistent with known environmental tolerances. Predicted changes in local biodiversity were associated with differences in both initial conditions and magnitude of alteration. However, the biota-flow associations were difficult to interpret because most flow metrics were correlated with temperature. Manipulative experiments will likely be needed to disentangle the effects of temperature and flow regime on stream biodiversity.