Abstract:Brain-Computer Interface (BCI) systems support direct communication and control between brain and external devices without use of peripheral nerves and muscles. A typical Electrocorticography (ECoG) based invasive BCI system was off-line analyzed in this paper. Firstly, Band Power (BP) features were used for channels selection, and 11 channels with distinctive features were selected from 64 channels. Then, BP features were used for feature extraction of 11 channels ECoG, and feature vectors of 22 dimensions were got. Finally, k Nearest Neighbor (kNN) was used for classification of two different mental tasks (imaged movement of left finger or tongue). The off-line analysis results show that this method has got good classification accuracy for the test data set.