P300　potential　is　weak　and　has　poor　anti-interference　ability　and　low　recognition　rate.　Based　on　wavelet　packet　transform　(WPT)　and　common　spatial　subspace　decomposition　(CSSD),　a　feature　extraction　method,　denoted　as　WPCSSD,　was　proposed　in　this　paper.　First,　the　EEG　was　preprocessed　by　the　overlapping　average　algorithm　to　improve　its　signal-to-noise　ratio.　Second,　the　EEG　was　filtered　and　reconstructed　by　WPT　according　to　the　time-frequency　information　of　P300.　Third,　the　power　spectrum　based　on　AR　model　was　computed,　and　a　spatial　filter　with　CSSD　was　applied　to　extract　the　spatial　feature　of　P300.　The　feature　vector　can　therefore　reflect　the　time-frequency-space　information　of　P300　generally.　Finally,　the　support　vector　machine　was　used　for　classification.　Results　show　that　WPCSSD　has　better　anti-interference　and　adaptive　ability,　and　the　recognition　accuracy　is　95.22%　in　data　sets　of　BCI　competition.　The　correctness　and　validity　of　the　method　are　proven.