Aimed　at　the　deficiency　of　soft-thresholding　denoising,　a　new　continuous　wavelet　transform　(CWT)　denoising　method　based　on　singular　value　decomposition　(SVD)　was　proposed.　The　matrix　of　CWT,　which　contains　signal　and　noise,　was　obtained　via　morlet　wavelet,　and　then,　instead　of　the　soft-thresholding　method,　the　SVD　was　applied　to　decompose　the　signal　features　and　noise　into　different　orthogonal　sub-spaces.　With　the　reconstruction　of　the　singular　vectors　in　sub-space,　the　signal　features　were　extracted　effectively.　Experimental　results　and　industrial　measurements　on　fault　diagnosis　show　that　the　approach,　compared　with　the　soft-thresholding　denoising,　is　an　efficient　tool　to　extract　impact　component　from　mechanical　signals.