P300　Speller　is　a　P300-based　Brain-Computer　Interfaces　(BCIs),　which　help　users　write　the　desired　characters　to　the　computer　screen　by　detecting　the　P300　event-related　potentials　in　users＇　electroencephalographic　(EEG).　This　article　proposes　a　feature　extraction　method　for　P300　detection.　This　method　combines　discrete　wavelet　transform　(DWT)　with　Fisher　criterion.　This　method,　firstly　transforms　the　EEG　signal　into　wavelet　domain　by　DWT.　Then,　the　feature　space　where　best　distinguish　two　kinds　of　EEG　signals　will　be　found　by　means　of　Fisher　criterion.　Finally,　a　feature　extraction　matrix　will　be　constructed　to　extract　feature　vector　from　EEG　signals.　The　method　was　tested　on　EEG　data　that　obtained　by　P300　Speller　paradigm.　The　result　shows　that,　the　proposed　method　is　better　than　the　existing　method　used　in　BCI2000,　in　terms　of　averaged　accuracy　over　238　runs.　©　2015　IEEE.