Abstract: Ultrasonic TOFD(time of flight diffraction) has the shortcoming of sensitive to noise for the lower amplitude in the tested signal. According to the characteristics that the defect waves have strong correlation and the noise non-correlation in TOFD tested two adjacent A-scan lines, an improved WMM(wavelet modular maximum) method based de-nosing is proposed. Computer simulated and weld tested signals are de-nosing processed by using conventional and improved WMM methods correspondingly. The experimental results show that the improved method is more effective in noise suppression and has better adaptability for different noise. Compared with the conventional method, the proposed technique can greatly improve the ability of identifying defect wave.