A non linear Model Predictive Controller (MPC) based on Artificial Neural Networks (ANN) is proposed for the control of a non linear open loop unstable SISO process. The system considered is an exothermic CSTR having multiple steady states. A non linear MPC is used for controlling the reactor at one of the steady states, which is unstable. For the desisn of non linear MPC. ANN is developed by obtaining input output data from the plant model equations. For this purpose a suitable artificial neural network (ANN) architecture is selected and is trained to compute one step-ahead predictions of plant's output. Further, a linear MPC based on the transfer function of the plant is developed. This controller stabilizes the linearized plant model but is unable to control the actual plant characterized by the non linear equations. In final analysis, results obtained for simulation of the non linear MPC and PID controllers demonstrate the superiority of non linear MPC control law. Copyright © IICAI 2005.