Abstract: Brain-Computer interaction(BCI) establishs a direct communication and control channel between human and computer or other electronic device by electroencephalogram (EEG). P300-based speller paradigm is an common communication between Brain and computer . The paper introduces a feature extraction method based on P300 visual evoked potential. Three subjects is used to analyse. Fixed point of independent component analysis (FastICA) and Fisher Discriminant Criterion are imploied to implement the feature extraction，and uses support vector machines (SVM) to classify EEG signal. Compared with the feature extraction based on PCA and Fisher Discriminant Criterion, it has a good ability to extract feature.