Since traditional Prony algorithm is extremely sensitive to noise, the Complementary Ensemble Empirical Mode Decomposition (CEEMD) and wavelet packet transform combined denoising method is used to improve the signal and enhance the Prony parameter identification accuracy. Firstly, the Intrinsic Mode Functions (IMF) are obtained through decomposing the signal with CEEMD method and their Permutation Entropy (PE) is calculated. Secondly, the high noise level IMFs which are extracted according to the values of the PE are processed with wavelet packet denoising. Thirdly, original signal is reconstructed by the IMFs which are processed and the rest IMFs. Finally, the parameters of reconstructed signal are identified with Prony algorithm. The algorithm proposed is simulated and compared with the literature results, the simulation and comparison results show that the algorithm is effective and has better identification results. This work is supported by Support Plan Foundation of Jiangsu Province (No. BE2013011-3).