In the process of ultrasonic testing, the quality of flaw echo signals in A-scan lay a better determinant foundation for location, quantitative and qualitative valuation. Flaw echo signal exited the characteristics of electrical noise, scattering noise in ultrasonic testing, which was sometimes very difficult to eliminate. Considering the distinctness of distribution between defects signals and noises, a time frequency neighborhood de-noise method based on wavelet packet transform was presented. Study mathematics properties of flaw echo signals and analyze the composition of noises and their characters in ultrasonic echo signals. According to the noise level and the variance in a neighborhood region, we presented an algorithm that wavelet transform coefficients of noise were filtered by changing threshold on the different scale and reconstructed the detection echo in order to enhance signal-to-noise ratio. Experiments with both simulated signals and measured signals have been carried out, and the result show that due to making the best use of the information of time and frequency domain at the same time in ultrasonic echo signals, this method can improve the signal noise ratio and the distinguish ability between signals of different defects classes, and suppress energy attenuation as well as signal distortion efficiently. And flaw location accuracy and longitudinal resolution are advanced too.