Abstract: In order to increase accuracy of the ultrasound non-destructive evaluation(NDE),it is required to reduce structural noise caused by randomly distributed scattering-grains in the material.Since spectra of the signal and noise overlap,traditional linear filtering methods cannot produce desirable de-noising results.In this paper,three filtering methods,i.e.,Wigner-Ville distribution,discrete wavelet transform and non-linear time-delay feed-forward dynamic neural network are studied.Three parameters,signal-to-noise ratio(SNR),probability of detection(PD) and estimated depth(ED) are calculated to compare the algorithm performance in the simulation studies.It is shown that wavelet transform and neural network perform better than Wigner-Ville distribution.Experiments also show that wavelet transform is an ideal de-noising technique for ultrasound NDE signals since it does not require a training process as used in neural networks.