With　the　development　of　Multimedia　Network　Technology　and　the　rapid　increase　of　image　application,　Content-based　Image　Retrieval　(CBIR)　becomes　the　most　active　one　in　multimedia　information　retrieval　field.　One　of　the　key　issues　in　CBIR　is　how　to　construct　effective　organization　and　index　to　enhance　image　retrieval　speed.　Clustering　is　a　kind　of　effective　method.　This　paper　presents　a　modified　fuzzy　C-means　(MFCM)　clustering　index　scheme　method.　In　addition,　in　order　to　reduce　the　time　of　clustering,　high-dimension　feature　space　is　transformed　into　lower-dimension　space　by　using　Karhunen-Loeve　(K-L)　transformation.　The　clustering　step　is　performed　in　lower-dimension　space,　and　image　retrieval　is　only　performed　in　clustered　prototypes.　Experimental　results　show　that　MFCM　applied　to　image　retrieval　is　effectively,　exact　and　real-time.　The　time　of　retrieval　doesn＇t　increase　linearly　with　the　extended　image　database.　It　is　superior　to　traditional　C-means　and　fuzzy　C-means　clustering　algorithms.