predictive models were developed to improve the understanding of stream-resident brook trout (salvelinus fontinalis) populations and habitat in northwestern ontario, and to facilitate protection of stream-resident brook trout from the adverse impacts of timber harvest. geology-based models correctly predicted trout presence/absence in 75%-80% of streams studied in 1993. however, correct prediction rates declined to 50%-65% when these models were transferred to independent data collected in 1992 and 1994. combining data from all years produced models that correctly predicted trout presence/absence in 70%-80% of streams. univariate geology models were best at predicting trout presence (up to 85% correct predictions). one-third of the trout streams data had maximum summer temperatures >22deg.c , and thus are considered marginal. using the combined data, models with geology and climate variables explained up to 24% of the variation associated with stream temperatures. stream temperatures were negatively related to brook trout abundance in the combined data. stability of stream temperatures accounted for 25% of the variation in trout biomass (kg/ha). these models could be used by fisheries managers to implement current guidelines protecting brook trout habitat from the effects of timber harvest.