The　traditional　Brain-computer　Interface　(BCI)　obtains　parameters　from　the　offline　analysis　and　applies　them　to　online　experiments.　However,　due　to　non-stationary　characte-ristic　of　electroencephalography　(EEG),　static　classification　of　algorithms　are　hard　to　be　used　in　practical　BCI.　In　this　paper,　we　propose　a　new　algorithm　that　combines　the　adaptation　of　preferable　new　incoming　data　with　the　incremental　linear　discriminant.　Then　we　design　a　new　experiment　paradigm　and　present　an　adaptive　BCI　algorithm　framework　to　meet　needs　of　the　online　experiment.　At　the　end,　we　look　for　8　health　subjects　to　participate　in　experiments　in　order　to　test　our　algorithm.　Results　of　our　experiments　showed　all　subjects　could　reach　60.7%　chance　level　to　the　final　session　and　the　best　result　was　65.7%　from　non-experienced　users　and　87.9%　from　people　with　experiences.　These　results　indicated　that　our　algorithm　is　effective.　In　addition,　we　discuss　the　difference　between　subjects　and　sessions　in　order　to　promote　accuracy　better　in　future.　We　consider　the　presented　online　experiment　method　is　the　first　step　that　towards　the　fully　autocalibrating　online　BCI　system.　©　2017　IEEE.