This work investigates the performance of a neural network (NN) for automatic generation control of a power system including the Superconducting Magnetic Energy Storage (SMES) unit. First, a single-area power system is represented by the state space model. The reheat effect nonlinearity of the steam turbine is also considered in this model. The control of this system is investigated using a conventional PIT controller versus an NN controller. Further computer simulations are performed on interconnected power system with two areas including SMES units. The single-area and two-area systems simulation results both indicate better control performance can be obtained by the NN controller. Only one NN controller, using the back propagation-through-time algorithm, controls the inputs of each area in the power system.