The purpose of this study was to demonstrate the effect of training on a subject's performance using an online Brain Computer Interface system with feedback. In order to do so, we developed an Online system BCILab which performs data acquisition, signal process- ing, feature extraction, classification and saving of all parameters. The data acquisition was done using a TCP connection with the ACQUIRE program. Spatial filter (common average reference) and detrending were used during signal processing to achieve high signal to noise ratio and stationarity of signal respectively.The power spectrum calculated for the autoregressive coeficients for channel C3 and C4 at 8, 10 and 12 Hz were used as the features for classification by a bayesian classifier. The data for training of the classifier was collected during a single screening session and multiple feedback sessions when the subject was asked to perform the left or right motor imagery task depending on target state. The performance of the subject was analysed over each session and over days. As the study had only two subjects, it was considered as a case study. However, in case of both subjects the performance of over days has shown improvement, in case of subject1 it improved from (48%±15%) to (67%±10%) and for subject2 (50%±0%) to (56%±11%) over 3 and 4 days respectively.