In recent studies, there have been attempts to control smart devices using Brain-Computer Interface (BCI) as a proof-of-concept. BCI can be a new interface between the human brain and various consumer electronic devices. Among various BCI paradigms, we adopted P300-based BCI which has higher accuracy with less variability among subjects and requires less training. We have explored the feasibility of TV channel selection based on BCI at typical home environment in terms of viewing distance (3 meters) and screen size (46 inches). Four candidate channel menus in video type were displayed on each quadrant of TV screen. On the upper-left or upper-right corner of each rectangular channel menu, visual stimulus was flashed to elicit Event Related Potentials (ERP). With 8 subjects, we achieved 92.3% average accuracy using 16 EEG channels and 89.6% average accuracy using only 3 EEG channels positioned on the midline sites of human brain. We found an optimal subset of EEG channels while keeping the performance.