AbstractStudies predicting how the distribution of aquatic organisms will shift with climate change often use projected increases in air temperature or water temperature. However, the assumed correlations between water temperature change and air temperature change can be problematic, especially for mountainous, high elevation streams. Using stream fish assemblage data from 1,442 surveys across a mountain–plains gradient (Wyoming, USA; 1990–2018), we compared the responsiveness of thermal guilds, native status groups, and assemblage structure to projected climate warming from generalized air temperature models and stream‐specific water temperature models. Air temperature models consistently predicted greater range shift differences between warm‐water and cold‐water species, with air temperatures predicting greater increases in occurrence and greater range expansions for warm‐water species. The “over‐prediction” of warm‐water species expansions resulted in air temperature models predicting higher rates of novel species combinations, greater increases in local species richness, and higher magnitudes of biotic homogenization compared with water temperature models. Despite differences in model predictions for warm‐water species, both air and water temperature models predicted that three cold‐water species would exhibit similar decreases in occurrence (decline of 1.0% and 1.8% of sites per 1°C warming, respectively) and similar range contractions (16.6 and 21.5 m elevation loss per 1°C warming, respectively). The “over‐prediction” for warm‐water species is partially attributable to water temperatures warming at slower rates than air temperatures because local, stream‐scale factors (e.g., riparian cover, groundwater inputs) buffer high elevation streams from rising air temperatures. Our study provides the first comparison of how inferences about climate‐induced biotic change at the species‐ and assemblage‐levels differ when modeling with generalized air temperatures versus stream‐specific water temperatures. We recommend that future studies use stream‐specific water temperature models, especially for mountainous, high elevation streams, to avoid the “over‐prediction” of biotic changes observed from air temperature variables.