In this paper we develop a new approach to the design of near-optimal decentralized controllers for large-scale linear interconnected dynamical systems. All calculations in the design stage are done on subsystems. This is achieved by using a simple reduced order model for the interactions. The resulting controller is decentralized, independent of initial conditions and capable of accommodating constant unknown disturbances. The approach is illustrated on a 22nd order river pollution control example. In addition, it is shown that by a simple modification it is possible to ensure that the controller is asymptotically stable with a pre-specified degree of stability and we show that the modified controller is also connectively asymptotically stable under structural perturbations.