Brain computer interface systems are modeled to facilitate lives of patients who have not a problem in their cognitive functions but also can not move their muscles. The performance of such systems highly depends on features extracted from the Electrocorticography (ECoG) signals, selected classifiers for features and channels of ECoG signals. In this study, we proposed a novel method which provides determination of effective ECoG channels in brain-computer interface applications. The proposed method not only increase the classification accuracy but also reduce the feature extraction time instead of using all the channels of recorded ECoG. The 92% classification accuracy rate was obtained by the proposed Sequential Forward Channel Selection algorithm. The achieved classification accuracy rate is 4% greater than the classification accuracy rate calculated by all channels. In addition, feature extraction time is reduced by 95.19% compared to feature extraction time using all channels.