To　improve　the　performance　of　support　vector　machines　(SVMs),　from　the　deep　learning＇s　point　of　view,　a　kernel　learning　method　was　studied　and　a　deep　kernel　mapping　support　vector　machine　(DKMSVM)　was　proposed　based　on　multi-layer　perceptron　together　with　the　corresponding　learning　algorithm.　Firstly,　a　kernel　mapping　from　the　original　input　space　to　a　proper　dimensional　space　through　a　multilayer　perceptron　instead　of　a　traditional　kernel　function　was　researched　in　this　model.　Then　a　SVM　was　used　to　classify　in　the　proper　dimensional　space　without　kernel　tricks.　Experimental　results　demonstrate　the　effectiveness　of　DKMSVM.　©　2016,　Editorial　Department　of　Journal　of　Beijing　University　of　Technology.　All　right　reserved.