Aquatic species throughout the interior Columbia River basin are at risk. Evaluation of the potential effects of federal land management on aquatic ecosystems across this region is an important but challenging task. Issues include the size and complexity of the systems, uncertainty in important processes and existing states,¯exibility and consistency in the analytical framework, and an ability to quantify results. We focused on salmonid ®shes and their habitats as indicators of conditions in aquatic ecosystems and used Bayesian belief networks as a formal, quantitative framework to address the issues in our evaluation of land management alternatives proposed for the interior Columbia River basin. Because empirical information is limited at the scales relevant to our analysis, an ability to combine both empirical and more subjective information was key to the analysis. The representation of linkages through conditional probabilities made uncertainty explicit. We constructed two general networks. One represented the in¯uence of landscape characteristics and existing and predicted management activities on aquatic habitats. A second represented the in¯uence of habitat, existing biotic conditions, and for two anadromous species, ocean and migratory conditions, on the status of six widely distributed salmonid ®shes. In the long term (100 years) all three land management alternatives were expected to produce positive changes in the status and distribution of the salmonids and their habitats. Trends were stronger for habitat than for the status of salmonids because of greater uncertainty in linking the ®sh and habitat networks and constraints outside spawning and rearing habitat on federal lands in the study area. Trends were stronger for resident salmonids than anadromous forms because of additional effects of the migratory corridor assumed for the latter. Alternative S2, which approached ecosystem restoration more conservatively, generally produced the strongest positive changes, and alternative S3, designed to promote more aggressive restoration, the weakest. Averaged across the basin, differences among the alternatives were small. Differences were greater at ®ner temporal and spatial scales. In the short term (10 years) alternative S3 was expected to lead to further degradation in some areas. By formalizing our understanding and assumptions in these networks, we provided a framework for exploring differences in the management alternatives that is more quanti®able, spatially explicit, and¯exible than previous approaches. Published by Elsevier Science B.V.