The　implementation　of　the　consistency　between　motor　intention　and　practical　rehabilitation　exercise　based　on　brain-computer　interface　technology　is　necessary　to　improve　the　rehabilitation　effect　for　people　with　dyskinesia.　Taking　the　flexion　and　extension　motor　imagery　of　index　finger　as　an　example,　the　feature　extraction　method　for　the　electroencephalogram　produced　by　the　same　or　similar　body　parts　under　different　motor　imagery　tasks　(labelled　as　EEGs)　is　studied　in　this　paper.　Aiming　at　the　characteristics　of　EEGs,　including　its　weak　phenomenon　of　event-related　desynchronization(ERD)　and　large　individual　differences　of　time　and　frequency　bands　where　ERD　appears,　an　optimal　frequency　band　extraction　method　is　proposed　based　on　wavelet　packet　decomposition　and　entropy　criterion.　The　EEGs　of　the　flexion　and　extension　motor　imagery　of　index　finger　are　decomposed　with　wavelet　packet　analysis　firstly.　Then,　the　separability　values　of　the　characteristic　frequency　bands　are　measured　with　entropy　criterion.　Furthermore,　some　clearer　wavelet　packets　are　selected　to　form　a　combination,　and　corresponding　wavelet　packet　coefficients　are　used　to　construct　the　feature　vectors.　Lastly,　the　optimal　band　is　obtained　with　support　vector　machine.　Experiment　results　show　that　the　feature　extraction　method　can　choose　the　feature　bands　with　large　difference　in　ERD　phenomenon　of　the　EEGs,　and　the　highest　classification　accuracy　is　81.75%,　which　verifies　the　correctness　and　validity　of　the　presented　method.