A　novel　analysis　method　is　proposed　based　on　ensemble　empirical　mode　decomposition　(EEMD)　and　support　vector　machines　(SVMs)　for　the　fault　diagnosis　of　bevel　gears.　Firstly,　the　EEMD　method　is　used　to　decompose　the　fluctuations　in　the　original　gear　noise　signals　into　different　timescales　so　as　to　obtain　several　intrinsic　mode　functions　(IMFs).　The　meshing　frequency　components　in　the　decomposition　results　are　reconstructed　to　eliminate　the　influence　of　interference　noise.　Then,　time-synchronous　averaging　(TSA)　is　applied　in　further　denoising　to　weaken　signals　independent　of　the　gear　meshing　frequency.　After　denoising,　various　signal　characteristics　are　calculated.　Obvious　signal　characteristics　for　different　fault　states　are　selected　as　a　set　of　feature　vectors.　Finally,　a　particle　optimisation　method　is　used　to　optimise　SVM　parameters　and　the　feature　vectors　are　input　as　training　samples　into　an　SVM　in　order　to　achieve　fault　recognition.　The　experimental　results　show　that　this　novel　analysis　method　can　effectively　diagnose　different　conditions　of　the　bevel　gear　and　achieve　an　identification　rate　for　gear　faults　of　98.33%.