Non-stationary　signals　are　a　mixture　of　sustained　oscillations　and　non-oscillatory　transients　that　are　difficult　to　analyze　by　linear　methods.　Aiming　at　this　problem,　a　nonlinear　signal　analysis　method　based　on　Q-factor　is　proposed,　which　expresses　the　non-stationary　signal　as　the　sum　of　a　high-resonance　(high　Q-factor)　and　a　low-resonance　component　(low　Q-factor).　And　then　the　dual　Q-factor　is　used　to　make　the　signal　be　sparse-decomposed　and　the　high-resonance　component　and　low-resonance　component　of　the　signal　are　obtained.　Applying　this　method　to　bearing　early　fault　diagnosis,　fault　signals　are　made　of　high-resonance　components　and　impulsion　fault　signals　(low-resonance　component).　And　impulsion　fault　signals　with　strong　background　noise　are　extracted　by　low　Q-factor　and　early　impact　characteristics　of　weak　damage　signals　of　the　bearing　are　extracted　successfully　and　quickly.　The　analysis　results　of　simulating　data　and　experimental　data　show　that　the　proposed　method　has　good　denoising　effect　and　it　could　remove　the　strong　background　noise　effectively.　©　2013　Journal　of　Mechanical　Engineering.