As　it　was　difficult　to　extract　weak　fault　feature　of　rolling　bearings　with　the　method　of　ensemble　empirical　mode　decomposition　(EEMD),　a　modified　EEMD　was　proposed.　First　the　method　of　minimum　entropy　deconvolution　(MED)　was　used　to　restrain　the　noise　and　highlight　the　impulse　components　of　vibration　signals.　Second　the　signals　were　decomposed　into　different　intrinsic　mode　function　(IMF)　by　using　EEMD,　and　then　the　sensitive　IMFS　were　selected　and　false　IMFS　were　eliminated　to　reconstructed　the　new　signals.　Third　the　noise　components　of　IMFS　were　restrained　by　the　method　of　wavelet-threshold.　The　power　spectrum　could　be　used　to　obtain　the　weak　fault　features　by　using　Teager　Energy　Operator　at　last.　The　diagnosis　results　with　the　simulation　signals　and　experimental　data　of　inner　race　faults,　had　indicated　the　effectiveness　and　accuracy　of　the　method.　©　2017,　Editorial　Department　of　Journal　of　Beijing　University　of　Technology.　All　right　reserved.