Electroencephalogram　(EEG)　is　a　comprehensive　indicator　of　human　physiological　activities.　Because　of　its　comprehensiveness　and　complexities,　an　electrode-covered　collection　device　on　the　scalp　cannot　collect　the　discharge　phenomenon　of　the　activated　brain　area　exhaustively.　A　single　electrode　analysis　will　miss　a　lot　of　important　association　information　between　different　brain　regions.　This　paper　presents　a　novel　strategy　to　solve　this　problem　by　combining　the　optimal　electrodes.　The　whole　method　is　divided　into　five　steps:　(1)　input　multi-electrodes　EEG　data,　(2)　utilize　the　Principal　Component　Analysis　(PCA)　method　to　obtain　the　optimal　electrodes,　(3)　adopt　the　modified　optimal　electrodes　recombination　strategy　and　obtain　optimal　electrode　combination　strategy,　(4)　extract　features　by　using　Empirical　Mode　Decomposition　(EMD),　and　(5)　use　Naive　Bayes　classifier　to　do　the　classification　tasks.　In　order　to　evaluate　the　validity　of　the　proposed　method,　we　apply　the　proposed　method　in　BCI　Competition　II　datasets　Ia.　Experimental　results　show　that　our　method　improves　recognition　performance　for　BCI　Competition　II　datasets　Ia　and　the　modified　optimal　electrodes　recombination　strategy　is　reasonable.　This　provides　a　new　idea　for　analyzing　the　EEG　features　of　other　tasks.