Recent　research　findings　in　brain　computer-interfacing　(BCI)　have　played　an　indispensable　role　for　stroke　patients　including　people　who　suffers　from　neural　injuries.　This　study　expects　to　better　quality　of　life　by　motor　imagery　and　replacement　of　neuro-muscular　pathways　as　BCI　systems　work　on　motor　imagery　to　control　prosthetic　limbs　for　rehabilitation　therapy.　In　this　research,　multiple　combinations　of　classifiers　have　been　used　to　classify　electroencephalographic　(EEG)　signals　in　order　to　acquire　maximum　classification　accuracy　for　a　three　classes　EEG　based　BCI　system.　Data　set　IVa　from　the　BCI　Competition　III　was　used　to　evaluate　the　performance　of　the　proposed　algorithm.　One-Verous-Rest　common　spatial　patterns(OVR-CSP)　is　commonly　used　to　extract　discriminative　spatial　filters　which　can　divide　the　three　types　of　motion　imaging　tasks　into　three　new　two-category　classification　tasks.　We　considered　using　OVR-CSP　to　extract　features　together　and　submitted　them　to　Naive　Bayes,　LDA,　QDA　and　SVM　algorithms　which　can　obviously　classify　left　hand,　right　hand,　and　right　foot　movement.　OVR-CSP　combined　with　LDA　classifier　performed　best　with　an　accuracy　of　84.62%.　Experiments　show　that　these　novel　findings　clearly　exhibit　the　feasibility　of　achieving　good　classification　accuracies　using　the　covariance　matrix　of　EEG　signals　as　feature　vectors　and　LDA　as　classifier.　Our　approach　presented　in　this　paper　is　relatively　simple,　easy　to　execute　and　is　validated　robustly　with　a　large　dataset.