Aimed　at　separating　fault　information　from　compound　rolling　bearing　fault　signal,　a　fault　diagnosis　method　was　proposed　based　on　dual-tree　complex　wavelet　packet　transform　and　auto-regressive　(AR)　spectrum.　First,　the　non-stationary　and　complex　signal　of　compound　fault　was　decomposed　into　several　different　frequency　band　components　through　dual-tree　complex　wavelet　packet　decomposition.　Second,　Hilbert　envelope　was　formed　from　the　component　that　contains　the　fault　information.　Finally,　the　power　spectrum　was　obtained　by　AR　spectrum.　Thus,　the　information　of　fault　feature　was　separated　and　identified.　Experiments　results　show　that　the　fault　feature　of　rolling　bearing　can　be　separated　effectively,　and　the　feasibility　and　effectiveness　of　the　method　are　verified.