The　article　put　forward　a　method　based　on　GM　(1,1)-SVM　for　rolling　bearing　fault　prediction　and　diagnosis.　Firstly,　the　method　extract　time　and　frequency　domain　feature　values　of　vibration　signal　of　rolling　bearing　under　all　kinds　of　fault　and　normal　condition.　Then　the　method　select　important　characteristic　parameters　to　build　a　grey　model　and　carry　on　multi　step　prediction;　Lastly,　the　method　use　all　kinds　of　fault　and　normal　condition　eigenvalue　to　train　binary　tree　support　vector　machine　and　construct　the　decision　tree　of　rolling　bearing　to　classify　the　fault　type.