A de-noising and recognition method based on HHT (HilbertHuang transform) was proposed to solve the problem that the traditional methods cannot effectively identify the pulse eddy current signals produced by small defects with different sizes. Firstly, the pulse eddy current signal was decomposed by EEMD (ensemble empirical mode decomposition), and the IMFs (intrinsic mode functions) of much noise were selected according to normalized autocorrelation function and its variance. Secondly, the selected IMFs of much noise were removed by the wavelet threshold denoising, and then the noiseless signal was reconstructed by adding to the nonprocessed IMFs. Then, the HMS (Hilbert marginal spectrum) was obtained by using HHT. Finally, according to the difference of HMS, the surface and subsurface defects with different sizes were identified. Experimental results show the effectiveness of the proposed method: the noise of pulsed eddy current signal is eliminated by noise elimination through EEMD, and the method based on HHT can effectively identify cracks of different sizes.