The　emotion　recognition　was　studied　by　using　the　entropy　analysis　of　EEG　signals,　and　an　algorithm　for　extraction　of　emotion　EEG　features　based　on　the　combination　of　permutation　entropy　and　multi　fractal　index　was　put　forward.　The　algorithm　achieves　EEG　feature　extraction　by　combinative　use　of　the　parameters　of　permutation　entropy,　Hurst　exponent,　mass　index　and　singular　spectrum　width,　and　achieves　the　emotion　recognition　by　using　Support　Vector　Machine　(SVM).　The　study　indicated　that　for　one-to-one　emotion　recognition,　the　highest　accuracy　of　the　testing　set　was　92.8%,　all　higher　than　80%　except　for　excitement　against　fear.　The　highest　accuracy　increased　by　41.9%　compared　with　the　permutation　entropy,　and　31.2%　compared　with　the　multi-fractal　index.　The　classification　effects　of　positive　emotion　and　passive　emotion　were　further　analyzed,　and　the　average　accuracy　of　test　set　was　78.3%,　respectively　increased　by　26.7%　and　1.6%　compared　with　the　entropy　and　the　multi-fractal　feature.　The　method　based　on　the　combination　of　permutation　entropy　and　multi-fractal　index　is　proved　to　be　an　effective　algorithm　for　emotion　EEG　feature　extraction,　with　the　capacity　of　sufficient　obtaining　the　nonlinear　trait　and　multi　fractal　feature　information.　©　2016,　Inst.　of　Scientific　and　Technical　Information　of　China.　All　right　reserved.