In this paper, we consider the decentralized robust stabilization problem of large-scale interconnected systems with delays, by making use of the theorem given by Lee and Radvic. If the mathematical system equation is complete, we can design decentlized control gains which satisfying the condition of the theorem so that the overall interconnected system is stable by using local state feedback. However, when the mathematical system equation includes model uncertainties, it is obvious that we can not make such interconncted systems stability by this controller. Thus, we propose the decentlized robust controller combining this controller into the multilayerd neural networks (MNN) to overcome system uncertainties.