The main steam temperature control system is necessary to ensure high efficiency and high load-following capability in the operation of modern power plant, which has the characteristics of large inertia, large time-delay and time-varying, etc. Thus conventional PID control strategy cannot achieve good control performance. The quadratic index was introduced into single-neuron PID controller and then the optimal controller was designed for accomplishing PID parameters' online adaptive optimization. This paper proposes a composite controller based on CMAC(Cerebellum Model Articulation Controller) Neural Network and the single-neuron PID controller with quadratic index, which the input of CMAC is the system's instruction signal, and taking the advantage of CMAC neural network with sample-structure, fast-convergence rate and the ability of local learning. A simulation study of the main steam temperature control system shows that this control strategy has the quality of strong robustness, adaptability and small overshoot.