This study analyses the location of patterns of brain activity in the signal space while a human subject is trained to operate a brain-computer interface. This evaluation plays an important role in the understanding of the underlying system, and it gives valuable information about the translation algorithms. The relative position and morphology of the patterns in a training session, and from one session to another, enable us to evaluate the performance of both the interface and the user. Thanks to these aforementioned variables we are also able to appreciate stable trajectories of the mental states during the sessions, which shows both the adaptability of the user to the interface, and vice versa.