A decentralized neural network controller design for large scale system with multiple uncertainties is presented in this paper. Controller synthesis relies primarily on an observer design, called the modified state observer (MSO). The uncertainties in the system are calculated online with the MSO. The MSO formulation of uncertainty estimation has two tunable gains which allow for fast estimation of the uncertainties and yet avoid high frequency oscillations that are usually observed with typical model reference adaptive controllers. Lyapunov analysis is used to show the stability of the system. Efficacy of the proposed controller is demonstrated using a bench mark numerical example.