According　to　the　widespread　problem　of　small　sample　learning　on　rolling　bearing　fault　diagnosis,　support　vector　machine　(SVM)　is　used　to　complete　the　pattern　recognition　of　bearing　fault.　In　order　to　solve　the　problem　of　poor　effect　of　multi-classification　bearing　faults　owing　to　time-domain　statistical　parameters,　the　Wavelet　Packet　Decomposition　(WPD)　technology　is　introduced　to　extract　energy　coefficient　of　each　vibration　signal　frequency　band　to　construct　feature　vector,　optimize　and　select　feature　vector　though　Fisher　ratio　method,　then　the　SVM　is　used　for　fault　pattern　recognition　and　comparative　analysis　of　the　classification　results　of　WPD　and　time-domain　statistical　parameters.　The　comparative　analysis　results　have　indicated　that　the　SVM　technology　is　an　effective　classification　method　for　fault　identification　of　rolling　bearings.　When　Fisher　ratio　method　combines　with　the　SVM,　the　fault　classification　accuracy　and　time　efficiency　is　higher　than　that　of　the　traditional　multidimensional　time-domain　and　WPD,　the　diagnosis　precision　can　also　be　improved.