Abstract: Denoising of ultrasonic testing signal is very important to analysis inner properties of tested materials. Matching pursuits decomposes any signal into a linear expansion of waveforms (atoms) that best match the signal structures. Then it isolates signal structures adaptively that coherent with respect to a given atoms dictionary and realizes denoising. In ultrasonic testing, the atoms is choosen as Gabor functions based on the model of ultrasonic echo signal, and the atoms can match the ultrasonic testing signals well. Simulated signal as well as experienced signal are used to testing the efficiency of matching pursuits denoising, and compared with wavelet transform denoising. The result shows that matching pursuits do a well job in denoising of ultrasonic noisy signals.