The　manifold　structure　analysis　of　face　image　components　in　RGB　color　space　was　given　in　the　paper.　A　novel　face　recognition　method,　which　integrates　manifold　learning　technique　with　the　color　information,　was　proposed.　In　this　method,　locally　linear　embedding　(LLE)　was　used　for　feature　extraction.　Afterward　normalization　and　fusion　was　done　for　the　extracted　features.　LDA　was　used　for　improving　classification　ability.　The　kNN　classifier　performed　face　recognition.　The　experiment　results　have　shown　that　the　proposed　method　can　improve　the　performance　of　both　the　Fishface　intensity　image　method　and　the　single　color　component　method　for　face　recognition.