Face　detection　is　important　in　the　processing　of　images　and　video.　Based　on　multilevel　gradient　energy　(MGE),　an　algorithm　of　face　detection　in　DCT　(Discrete　Cosine　Transform)　compressed　domain　is　presented.　In　preprocessing　procedure,　skin　color　segmentation　based　on　the　DC　of　chromatic　components　is　applied　to　the　input　image　for　reducing　the　detected　regions.　According　to　the　map　of　skin　segmentation,　MGE　based　feature　vector　is　extracted,　viz.　normalized　feature　vectors　are　extracted　from　the　detecting　windows　of　various　sizes　to　describe　faces　of　different　sizes.　Then　cascade　classifier　is　employed　to　classify　the　feature　vectors　as　face　or　non-face.　Cascade　classifier　is　comprised　of　several　simple　classifiers　and　a　neural　network　classifier.　Lots　of　feature　vectors　that　belong　to　non-face　are　removed　by　simple　classifiers　which　embedded　preknowledge　rules.　The　left　vectors　are　classified　by　neural　network.　We　combined　MGE　features　together　with　image　scaling　to　allow　faces　of　various　sizes.　The　simplicity　of　feature　extraction　accelerated　detection　by　reducing　the　times　of　image　scaling　which　is　more　time　cost.　The　experiment　results　show　that　the　proposed　method　is　efficient　and　effective.