Abstract: In ultrasonic TOFD(time of flight diffraction) tested signal, the phase information is very important parameter for qualitative identification and quantitative measurement of defect usually, Phase characteristic is not obvious in the original tested signal for the effect of testing system. For low SNR(signal to noise ratio) ultrasonic signal, the Wiener inverse filter is insufficient for improving the time resolution. To solve the disadvantuge of traditional de-convolution technique, an improved Wiener inverse filter method is proposed by employing spectral extrapolation. Computer simulated and artificial defect tested signals were processed by using traditional and improved Wiener filter technology respetively. The results showed that the proposed method could effectively improve the time resolution of the received signal, and could accurately extract the phase information about the defect tips. The proposed method could provide a reliable basis for defect identification and location.