Most reported applications of model-predictive control (MPC) have a narrow scope, with 1-5 variables to be regulated and a comparable number of manipulated variables. Several authors have claimed that global-optimization versions of MPC (such as DMC and QDMC) should be more useful for problems in which an entire system can be operated to achieve an economic and/or technical objective. In this paper, we describe the application of MPC to a large-scale, constraint-dominated problem : the minimization of combined-sewer overflows (CSOs) in the Seattle metropolitan area. The key decision variables are flowrates at 23 locations throughout the sewer network. There are approximately 40 output variables that must be kept between lower and upper bounds. The main issues addressed in the application are : (1) definition of an appropriate objective function for on-line optimization; (2) creation and maintenance of complex system models; and (3) use of state estimation to minimize the impact of disturbances and model errors. MPC performance is compared with that of an existing heuristic (rule-based) control strategy for seven design storms, selected from historical records. A realistic, nonlinear simulation of the sewer system acts as the plant. MPC reduces CSOs by 26% (on a yearly basis) relative to the existing control strategy. This was sufficient incentive for the sewer agency to replace their heuristic control strategy with MPC.