During the training, the user is given a feedback that allows him/her to see whether or not the system recognised the right task but also how well it could recognise it. By create a computational model of training that could predict the performances of a BCI user based on the feedback characteristics such as its strength or its frequency we would be able to identify other types of factors to be considered in order to improve training. We are currenlty building models of performance based on feedback characteristics such as the stability of the feedback bar, the duration of visibility by the user or it strength.