Compound　fault　feature　separation　is　a　difficult　problem　in　diagnosis　field　of　mechanical　system.　For　the　rolling　bearing　with　compound　fault　on　outer　and　inner　race,　feature　separation　technology　based　on　complex　wavelet　transform　and　energy　operator　demodulation　is　introduced.　Through　continuous　wavelet　transform,　coefficients　of　mixed　fault　signal　can　be　achieved　in　different　wavelet　transform　domain　(i.e.　real,　imaginary,　modulus　and　phase　domain).　Furthermore,　wavelet　power　spectrum　contours　and　time　average　wavelet　energy　spectrum　are　applied　to　extract　the　scales　which　hold　rich　fault　information,　and　the　wavelet　coefficient　slice　of　specific　scale　is　also　drawn.　For　wavelet　coefficients　in　different　domain,　spectrum　analysis　and　energy　operator　demodulation　can　be　used　successfully　to　separate　mixed　fault.　The　comparison　of　feature　extraction　effect　between　complex　wavelet　and　real　wavelet　transform　shows　that　complex　wavelet　transform　is　obviously　better　than　the　latter.