A　gear　fault　detection　analysis　method　based　on　Fractional　Wavelet　Transform　(FRWT)　and　Back　Propagation　Neural　Network　(BPNN)　is　proposed.　Taking　the　changing　order　as　the　variable,　the　optimal　order　of　gear　vibration　signals　is　determined　by　discrete　fractional　Fourier　transform.　Under　the　optimal　order,　the　fractional　wavelet　transform　is　applied　to　eliminate　noise　from　gear　vibration　signals.　In　this　way,　useful　components　of　vibration　signals　can　be　successfully　separated　from　background　noise.　Then,　a　set　of　feature　vectors　obtained　by　calculating　the　characteristic　parameters　for　the　de-noised　signals　are　used　to　characterize　the　gear　vibration　features.　Finally,　the　feature　vectors　are　divided　into　two　groups,　including　training　samples　and　testing　samples,　which　are　input　into　the　BPNN　for　learning　and　classification.　Experimental　results　showed　that　this　gear　fault　detection　analysis　method　could　well　maintain　the　useful　signal　components　related　to　gear　faults　and　effectively　extract　the　weak　fault　feature.　The　accuracy　rate　reached　96.67%　in　the　identification　of　the　type　of　gear　fault.