In order to control the turbojet engine in the all operation condition and whole flight envelope, this paper establishes a new self-adaptive neural network control system which is conducted by a neural network controller similar to PID and an identifier based on the Elman neural network with the self-feedback. Elman network is adopted as plant model predictor to identify controlled plant on-line. Conjugated gradient descent method is used to speed the convergence. Massive stimulations to the pilotless aircraft turbojet engine with MATLAB prove several advantages of this controlling method, such as fitting for whole flight envelope control, strong robustness, swift response, and minimal steady-stable error.