Skin　color　segmentation　is　a　fundamental　approach　in　the　domain　of　face　detection　of　color　images.　An　algorithm　of　skin　color　segmentation　applied　in　the　DCT　compressed　domain　is　proposed.　The　3D　ellipse　model　is　used　to　characterize　the　statistic　feature　of　skin　color　in　YCbCr　space.　Research　has　shown　that　luminance　has　a　non-linear　effect　on　the　cluster　of　skin　colors　in　the　chromatic　plane.　Different　ellipses　are　used　to　describe　the　cluster　of　skin　colors　with　different　luminances,　and　the　image　is　segmented　by　this　model　using　DC　(Direct　Current)　coefficients.　On　the　observation　that　the　texture　feature　of　the　skin　color　area　is　similar　in　the　same　image,　the　segmentation　result　is　refined　using　AC　(Alternating　Current)　coefficients.　Background　noise　which　is　left　in　the　first　segmentation　is　removed　by　adaptive　segmentation.　Experimental　results　show　that　the　proposed　algorithm　is　an　effective　skin　color　segmentation　method　in　the　compressed　domain.