Co-adaptation between the human brain and computers is an important issue in brain-computer interface (BCI) research. However, most of the research has focused on the computer side of BCI, such as developing powerful machine-learning algorithms, while less research has focused on investigating how BCI users may optimally adapt. This paper assesses the influences of positive and negative visual feedback on motor imagery (MI) skills by evaluating the performance. More precisely, a MI based BCI paradigm was employed with fake visual feedback, regardless of subjects' real performance. Subjects were exposed to two experimental conditions--one positive and one negative, in which 80% or 30% of the trials were associated with positive feedback, respectively. The main EEG feature for MI-BCI classification--the asymmetry of mu-rhythm between hemispheres--was more prominent only after the negative feedback session. In addition, the negative feedback condition was accompanied by larger heart rate variability compared to the positive feedback condition. Our results suggest that visual feedback is an important aspect to take into account when designing BCI skill acquisition sessions.