The　Latent　Semantic　Indexing　based　standard　retrieval　usually　computes　semantic　similarity　between　the　query　and　every　document,　which　is　sensitive　to　the　noise.　Furthermore,　it　will　take　much　more　time　when　the　scale　of　the　corpus　becomes　larger.　We　propose　a　retrieval　method　which　is　based　on　an　indexing　matrix.　The　noises　can　be　wiped　off　at　a　certain　extent　and　it　can　get　a　better　result　than　the　traditional　ways.　The　time　cost　of　our　method　is　much　less　than　the　standard　retrieval　method.　©　2008　Springer-Verlag　Berlin　Heidelberg.