Abstract: To solve the problem of vibration signals acquired from flue gas turbine interfered by noise, the principle and method of wavelet packet-based penalty function soft-threshold de-noising are analyzed and its procedure of de-noising is provided. The rule of Birgé-Massart penalty function to determine threshold is clarified and soft-threshold quantization processing is made. Then, the value changes of threshold, SNR and mean square error with the penalty factor are observed and compared with Rigrsure, Heursure, Sqtwolog, Minimax soft-threshold de-noising methods. The results show that the penalty function soft-threshold method can effectively reduce the noise. The de-noised vibration signals of flue gas turbine based on this method can retain mutant part of the signal, while it still has good smoothness.