The　traditional　technology　of　JPEG　image　steganography　blind　detection　is　implemented　by　a　single　feature　or　the　fusion　of　two　features.　There　has　been　no　published　research　yet　considering　how　to　select　a　superior　feature　combination　to　achieve　better　detection　performance.　Here　we　get　ideas　from　Hoffmann　trees,　put　forward　a　superior　feature　combination　selection　method　based　on　treelike　structure,　which　selects　correlation　coefficient　of　canonical　correlation　analysis　as　its　heuristic　factor.　In　the　method　we　try　to　fuse　two　feature　combinations　which　have　the　least　correlation　coefficient　every　time,　and　remain　the　feature　combination　with　higher　accuracy　rate.　Finally,　the　superior　feature　combination　can　be　obtained　in　the　way.　Experiments　indicate　that　our　method　can　get　a　feature　combination　which　holds　better　fusion　performance　in　limited　computing　complexity　condition　when　we　deal　with　steganography　blind　detection　for　JPEG　images.　©　2012　IEEE.