Application of stream water-quality models in decision making has been hampered by a lack of data appropriate for minimization of model-simulation uncertainty. A method for determining data needed to reduce model-prediction uncertainty is illustrated in this paper. First-order reliability analysis is applied to determine (1) the model parameters that significantly affect model-prediction uncertainty; and (2) the constituents for which model-prediction uncertainty is unacceptable. Additional data are required to reduce uncertainty in the parameters that significantly affect constituents with high prediction uncertainty and consequently in model prediction. The method is demonstrated for multiconstituent water-quality modeling on the Passaic River in New Jersey applying QUAL2E. The model-prediction uncertainty of dissolved oxygen, biochemical oxygen demand, ammonia, and chlorphyll a is considered. For this example, only the reaeration rate and the algal maximum-specific-growth rate contribute significant uncertainty to model prediction. The effect of reducing the uncertainty in the reaeration rate and algal maximum-specific-growth rate on the uncertainty on predicted dissolved oxygen and chlorphyll a, respectively, is demonstrated.