The　vibration　signals　of　transformer　oil　tank　can　represent　the　fault　of　inner　winding　and　core,　which　has　pretty　good　application　prospect.　But　the　signal　measured　on　the　oil　tank　surface　is　not　only　mixture　of　winding　and　core　vibration　but　also　dependent　with　each　other,　therefore,　it　is　difficult　to　separate　with　normal　BSS　arithmetic.　As　a　result,　it＇s　hard　to　diagnose　the　fault　of　winding　and　core　respectively.　This　paper　introduced　a　method　based　on　Sub-band　decomposition　independent　component　analysis(SDICA)　arithmetic　to　separate　winding　and　core　vibration　signal.　First,　SDICA　arithmetic　which　can　separate　the　signals　with　a　certain　correlation　degree　was　introduced.　Then　simulated　vibration　signals　were　adopted　to　test　and　compare　the　ability　of　SDICA　with　fast　independent　component　analysis(fastICA)　in　separating　mixture　signals.　The　results　show　that　SDICA　arithmetic　is　valid　to　separate　transformer　vibration　signals.　Then　applied　the　SDICA　arithmetic　to　the　vibration　signals　measured　from　a　test　transformer　and　discussed　the　effect　of　measuring　position,　voltage　and　load　to　the　separate　result.　Finally,　applied　the　SDICA　arithmetic　to　the　vibration　signals　measured　from　a　power　transformer　which　running　with　a　hidden　danger.　The　results　show　that　winding　and　core　vibration　signals　are　separated　successfully　and　the　fault　signature　can　be　seen　clearly　from　it,　which　is　correspondence　to　the　failure　condition.　It　can　be　drawn　a　conclusion　that　the　SDICA　arithmetic　introduced　in　this　paper　is　robust　in　separating　the　vibration　signals　measured　from　transformer　oil　tank.　The　results　are　significant　in　extending　the　vibration　method　to　the　fault　diagnosing　in　transformer　winding　and　core.