Abstract: Signal processing in ultrasonic test of austenitic welds was studied.Based on time-frequency analysis,a comprehensive signal processing method was proposed,which was a combination of matching pursuit and wavelet analysis.This method distinguishes flaw information and material noise from a point of energy and frequency,respectively,and improves the SNR of the ultrasonic echo signal.To verify the effectiveness of the method,an ultrasonic test system was established.Test experiment was carried out with a 53 mm thick weld specimen,and a signal with large amount of material noise was acquired.The noised signal was processed by the proposed method.The first 18 time-frequency atoms were extracted by matching pursuit with sym8 wavelet packet dictionary,the extracted signal was analyzed by sym8 wavelet decomposition and reconstructed with low frequency coefficients.The result shows that the material noise is reduced effectively and the flaw echo signal is enhanced significantly.