Assessing ecosystem health on large rivers is often complicated because these ecosystems receive multiple, interacting effluent discharges. Confounding factors, such as complicated mixing hydraulics and historical loading effects, can result in equivocal field data that lend weak inference to ecological risk assessments. Our approach to this problem develops a strategy that defines important mechanisms of pollutant effects through the combined use of laboratory and field measurements, riverside mesocosm experiments, and the incorporation of indicators at several trophic levels. We integrate these different types of information through weight-of-evidence postulates that provide logical guidelines for establishing causation in ecological risk assessment. Using this approach, retrospective risk assessments on the Fraser River, British Columbia, Canada, indicated that the major effect of present effluent discharges has been one of nutrient enrichment and stimulation of food web productivity. In fact, the Fraser River study suggests that small increases in effluent concentration in the river may produce negative ecological effects because of contaminant stresses. The combination of field experiments with this weight-of-evidence approach yielded the scientific justification for a conceptual model that describes community shifts across a nutrient-contaminant gradient. We conclude that the use of stream mesocosms with the weight-of-evidence approach is highly useful for establishing a mechanistic understanding of community responses to stressors at a regional scale. In addition, this approach will be useful when greater understanding of a particular class of anthropogenic stressors (e.g., specific types of effluent) is required to improve regulatory guidelines.