In　this　paper,　an　empirical　mode　decomposition　(EMD)　based　approach　for　rotating　machine　fault　diagnosis　is　investigated.　EMD　is　a　new　time-frequency　analyzing　method　for　nonlinear　and　non-stationary　signals.　By　using　EMD　a　complicated　signal　can　be　decomposed　into　a　number　of　intrinsic　mode　functions　(IMFs)　based　on　the　local　characteristic　time　scale　of　the　signal.　The　IMFs,　working　as　the　basis　functions,　represent　the　intrinsic　oscillation　modes　embedded　in　the　signal.　However,　our　research　shows　that　IMFs　sometimes　fail　to　reveal　the　signal　characteristics　due　to　the　effect　of　noises.　Hence,　combined　mode　function　(CMF)　is　presented.　With　CMF,　the　neighboring　IMFs　are　combined　to　obtain　an　oscillation　mode　depicting　signal　features　more　precisely.　The　adaptive　filtering　features　of　EMD　and　CMF　are　discussed,　and　the　simulation　signals　are　applied　to　test　their　performance.　Finally,　a　practical　fault　signal　of　a　power　generator　from　a　thermal-electric　plant　is　analyzed　to　diagnose　the　fault　by　using　EMD　and　CMF.　The　results　show　that　EMD　and　CHF　can　extract　the　rotating　machine　fault　characteristics　and　identify　the　fault　patterns　effectively.　(c)　2007　Elsevier　Ltd.　All　rights　reserved.