The studies in brain computer interface field generally deal with about either improving the performance of their works or investigating new applications. In this study, it is aimed to improve the effectiveness of common spatial pattern features by determining the optimum duration of electroencephalogram based brain computer interface trials. With the proposed method it is shown that the classification accuracies are increased in between 20.62% and 4.83%, and the determination time are accelerated by decreasing the trial durations.