Wavelet　thresholding　as　a　common　signal　processing　method　for　fault　diagnosis　of　rolling　bearing　has　the　deficiency　of　difficulty　to　choose　basic　function　and　the　weakness　of　poor　denoising　performance　by　using　conventional　soft　or　hard　threshold.　A　method　combining　EMD　interval-thresholding　with　maximum　likelihood　estimation　to　diagnose　the　incipient　weak　fault　of　rolling　bearing　was　presented.　The　original　signal　was　analyzed　by　empirical　mode　decomposition　(EMD),　then　each　intrinsic　modal　function　(IMF)　was　denoised　by　interval-thresholding　based　on　maximum　likelihood　estimation　and　the　fault　signals　was　acquired　by　reconstructing　the　thresholded　IMFs.　Finally,　the　results　were　achieved　by　envelope　spectrum　analysis　of　denoised　signal.　The　results　of　numerical　simulation　and　an　industrial　case　show　that　the　proposed　method　is　effective　to　diagnose　the　fault　of　rolling　bearing　significantly.