Temperature is an important variable structuring lotic biotas, but little is known about how montane landscapes function to determine stream temperatures. We developed an a priori hypothesis that was used to predict how watershed elements would interact to affect stream temperatures. The hypothesis was tested in a series of path analyses using temperature data from 26 sites on second-order to fourth-order streams across a fifth-order Rocky Mountain watershed. Based on the performance of the first hypothesis, two revised versions of the hypothesis were developed and tested that proved to be more accurate than the original hypothesis. The most plausible of the revised hypotheses accounted for 82 percent of the variation in maximum stream temperature, had a predicted data structure that did not deviate from the empirical data structure, and was the most parsimonious. The final working hypothesis suggested that stream temperature maxima were directly controlled by a large negative effect from mean basin elevation (direct effect = -0.57, p < 0.01) and smaller effects from riparian tree abundance (direct effect = -0.28, p = 0.03), and cattle density (direct effect = 0.24, p = 0.05). Watershed slope, valley constraint, and the abundance of grass across a watershed also affected temperature maxima, but these effects were indirect and mediated through cattle density and riparian trees. Three variables included in the a priori hypothesis - watershed aspect, stream width, and watershed size - had negligible effects on maximum stream temperatures and were omitted from the final working hypothesis.