Aiming　at　the　reduction　of　strong　background　noise　involved　in　non-stationary　signals　of　fault　gear　and　the　difficulty　to　obtain　fault　frequencies　in　practice,　a　new　fault　diagnosis　method　was　proposed　based　on　dual-tree　complex　wavelet　transform　and　singular　value　difference　spectrum.　Original　fault　signals　were　decomposed　into　several　different　frequency　band　components　through　dual-tree　complex　wavelet　decomposition.　But　it　is　often　difficult　to　obtain　the　exact　fault　frequencies　from　the　components　because　of　the　existence　of　strong　background　noise.　A　Hankel　matrix　was　constructed　making　use　of　the　component　which　contains　the　fault　information,　and　the　singular　value　difference　spectrum　was　obtained　after　the　singular　value　decomposition.　Then　the　number　of　singular　values　was　obtained　to　realize　signal　de-noising　by　the　SVD　reconstruction.　Finally,　the　fault　frequency　can　be　identified　accurately　by　Hilbert　envelope　spectrum.　The　results　of　the　experiments　and　engineering　applications　show　that　the　fault　feature　of　gear　can　be　extracted　effectively.　The　feasibility　and　effectiveness　of　the　method　were　verified.