Recently,　multi-dimensional　support　vector　regression　(M-SVR)　has　been　a　promising　tool　in　solving　a　wide　range　of　multi-input　and　multi-output　(MIMO)　regression　problems.　However,　when　some　output　dimensions　have　more　dependencies　than　others,　M-SVR　is　are　generally　hard　to　obtain　impressive　performance　due　to　the　negative　or　redundant　domain　knowledge　across　all　outputs.　A　structure　of　adaptive　grouping　is　introduced　into　the　M-SVR　to　solve　the　problem.　It　is　assumed　that　model　parameters　of　related　output　dimensions　are　similar　to　each　other,　then　a　new　regularization-based　M-SVR　algorithm　is　presented　by　introducing　a　regularizer　with　grouping　structure.　The　regularization　problem　is　then　converted　into　a　mixed　0-1　programming　problem.　Alternating　optimization　is　employed　for　learning　related　outputs　jointly　in　the　same　group,　and　to　obtain　the　optimal　grouping　structure　and　model　parameters　through　adaptively　identifying　the　grouping　structure.　The　proposed　algorithm　is　tested　empirically　on　a　toy　problem　as　well　as　a　real-life　data　set.　The　experimental　results　show　that　the　performance　of　the　proposed　algorithm　outperforms　the　performance　of　the　classical　M-SVR　and　the　single-output　SVM　algorithm.