Traditional chance-constrained programming (CCP) and simulation-optimization methods of incorporating input information uncertainty in pollution management models are unsuitable for complex river systems with several critical water quality segments. Using the CCP method, characterization of the joint probability distribution of coefficients of the management model is often difficult because stream information is limited and the model formulation is generally difficult to understand and solve. For the simulation-optimization method most of the solutions produced are inferior. The multiple realization model, which includes several scenarios of design conditions simultaneously in an optimization model, overcomes such weaknesses by not requiring the joint probability distribution of the stochastic model coefficients and by producing noninferior solutions. Heuristic and neural network techniques are developed to reduce the computational time required to solve the multiple realization model, through identification and utilization of only potentially important stream and water quality information that influence the optimal solution. These techniques are applied to develop trade-off relationships between waste treatment cost and reliability of achieving dissolved oxygen objectives for an example river basin. Results show that the heuristic technique is computationally efficient when <1000 realizations are included in the model, while the neural network method is suitable when several thousand realizations are needed to adequately represent the stochastic water quality system. ¿ 1999 American Geophysical Union