Abstract: A modulation classification scheme based on the chaotic theory is presented in low SNR condition by taking the advantage of the chaotic oscillator at noise management. The principle and method of the modulation classification to MPSK signal are given, and a new Duffing oscillator model is designed to improve the classification performance. The method only uses the prior information of carrier frequency, so it can be implemented without the parameter estimation of phase offset, symbol rate and noise power. Simulation experiments show that the probability of correct classification is larger than 95% for classifying 3 kinds of MPSK signals respectively when the SNR is as low as -5 dB. It verifies the validity of the algorithm.