Lattice　reduction　technology　has　been　proved　to　improve　MIMO　system　performance　recently,　but　the　complexity　of　optimum　quantization　algorithm　is　high　due　to　constellation　biased　distribution　caused　by　signal　transformation.　In　this　paper,　a　sorted　feedback　non-linear　quantization　(SFQ)　scheme　is　proposed,　which　considers　both　the　border　of　transformed　constellation　and　the　correlations　between　signal　elements,　and　the　elements　are　selected　to　be　quantized　according　to　quantization　errors.　The　simulation　results　show　that　the　performance　of　lattice　reduction　aided　(LRA)　MIMO　detectors　based　on　the　proposed　quantization　scheme　is　superior　to　that　of　LRA　MIMO　detectors　based　on　non-sorted　feedback　and　independent　element-wise　quantization　schemes.　The　proposed　algorithm　can　approach　the　performance　of　maximum　likelihood　detection　in　polynomial　complexity　in　a　flat　block　fading　channel.　©　2010　IEEE.