The　study　on　multi-scale　entropy　applied　to　extract　emotional　EEG　features　has　been　researched.　Considering　that　the　traditional　feature　extraction　algorithm　adopting　multi-scale　entropy　may　lose　important　information　during　the　coarsening　process,　and　that　small-scale　suffers　from　lack　of　significant　feature　while　big-scale　causes　excessive　computation,　an　improved　extraction　algorithm　based　on　multi-scale　entropy　was　put　forward.　The　improved　algorithm　determines　scales　according　to　the　number　of　intrinsic　mode　functions　of　the　adaptive　multi-scale　entropy.　At　the　same　time,　the　EEG　signal＇s　small　changes　were　highlighted　through　the　binary-state　processing　of　the　EEG　signal　with　the　adaptive　method,　thus　the　characteristics　of　the　EEG　signal　can　be　fully　tapped,　and　the　complexity　of　the　algorithm　can　be　reduced.　Based　on　the　optimized　Support　Vector　Machine　(SVM)　classifier,　the　emotional　EEG　recognition　was　achieved　by　using　the　international　DEAP　database　for　emotion　analysis,　and　the　performance　of　the　improved　algorithm　was　compared　with　traditional　algorithm＇s.　The　results　indicate　that　the　classification　accuracy　of　the　improved　algorithm　is　12.33%　higher　than　that　of　the　traditional　multi-scale　entropy　algorithm　and　7.27%　higher　than　that　of　the　adaptive　multi-scale　entropy　algorithm,　showing　that　the　improved　algorithm　is　effective　in　extracting　EEG　features.