The　traditional　singular　value　decomposition　(SVD)　method　is　unable　to　diagnose　the　weak　fault　feature　of　bearings　effectively,　which　means,　it　is　difficult　to　retain　the　effective　singular　components　(SCs).　Therefore,　a　new　singular　value　decomposition　method,　SVD　based　on　the　FIC　(fault　information　content),　is　proposed,　which　takes　the　amplitude　characteristics　of　fault　feature　frequency　as　the　selection　index　FIC　of　singular　components.　Firstly,　the　Hankel　matrix　of　the　original　signal　is　constructed,　and　SVD　is　applied　in　the　matrix.　Secondly,　the　proposed　index　FIC　is　used　to　evaluate　the　information　of　the　decomposed　SCs.　Finally,　the　SCs　with　fault　information　are　selected　and　added　to　obtain　the　denoised　signal.　The　results　of　bearing　fault　simulation　signals　and　experimental　signals　show　that　compared　with　the　traditional　differential　singular　value　decomposition　(DS-SVD),　the　proposed　method　can　select　the　singular　components　with　larger　amount　of　fault　information　and　is　able　to　diagnose　the　fault　under　the　heavy　noise　interference.　The　new　method　can　be　used　for　signal　denoising　and　weak　fault　feature　extraction.