This paper presents a decentralized Model Predictive Control (MPC) for Automatic Generation Control (AGC) of a realistic multi power system in Egypt with inherent nonlinearities. The Egyptian Power System (EPS) is decomposed into three dynamic subsystems which are non-reheat, reheat, and hydro power plants. Moreover, each subsystem has its private characteristics compared to the others. Therefore, each subsystem controller has been designed independently to guarantee the stability of the overall closed loop system. Hence, MPC is proposed for every subsystem separately to regulate the frequency and to track the load demands of EPS. The performance of the proposed decentralized MPC of each subsystem is compared with the centralized one at different operational scenarios. The results by nonlinear simulation MATLAB/SIMULINK for the generation control of EPS approves that the decentralized predictive model gives almost the same performance as the centralized one. However, the centralized model cannot address the large load disturbance and system parameters variation. In contrast, the decentralized MPC scheme is more robust and effective against all disturbances and operating conditions.