To　remove　electromyography　(EMG)　artifacts　from　electroencephalogram　(EEG)　signals　real-time,　canonical　correlation　analysis　(CCA)　is　adopted.　By　analyzing　a　number　of　＇clean＇　and　contaminated　electroencephalogram　(EEG)　signals　using　CCA,　a　reasonable　correlation　threshold　is　obtained.　The　EMG　artifacts　are　similar　to　the　common　noise　in　time　domain.　Hence,　the　EMG　artifacts　components　obtained　by　CCA　have　relatively　lower　correlation　than　non-EMG　artifacts.　When　CCA　is　used　to　remove　EMG　artifacts　from　EEG　signals,　the　components　whose　correlation　value　is　lower　than　the　threshold　are　identified　as　EMG　artifacts,　and　then　the　＇clean＇　EEG　signals　can　be　reconstructed　by　the　remnant　components.　The　experimental　results　show　that　CCA　outperforms　ICA　for　removing　EMG　artifacts.　Moreover,　combining　with　the　presented　threshold,　CCA　enables　to　effectively　remove　EMG　artifacts　with　little　distortion　of　the　underlying　brain　activity　signals　in　real-time.