A brain-computer interface (BCI) is a nonmuscular communication channel which allows physically disabled people to re-establish interaction with their surrounding environment. The aim of this system is to detect user's intentions or thoughts from brain activities which are measured as EEG (electroencephalogram) or other non-invasive recording techniques. We have developed the BCI system to detect user's motor imagery from motor-related EEG activities. It was shown that online training with a feedback of EEG band-power on specified frequency range made the accuracy of command detection higher. And we proposed an auditory BCI based on auditory stream segregation, on which users were requested to attend to one of the tone streams and their object of interest was detected from auditory event-related potentials. In this article, the concept and some of our recent researches on BCI are introduced.