The　facial　texture　variation　is　the　key　factor　for　realistic　expression　synthesis.　It　always　changed　with　facial　motion　under　some　illumination.　In　this　paper,　we　propose　a　realistic　facial　expression　texture　driving　model　based　on　the　support　vector　regression　and　MPEG-4.　It　can　learn　and　recall　the　regression　relationship　between　facial　animation　parameters　and　the　parameters　of　expression　ratio　image　through　support　vector　regression　method.　First,　We　can　get　the　parameter　set　of　expression　ratio　image　and　the　eigenERI　space　by　principle　component　analysis　method,　who　will　generate　reasonable　ratio　image.　Then,　a　life-like　facial　animation　can　be　synthesized　quickly　and　effectively　with　the　support　vector　regression　mapping.　In　our　experiment,　it　not　only　captures　subtle　changes　in　the　variation　illumination,　but　also　can　synthesis　realistic　facial　expression　in　bad　environment.　©　2004　IEEE.