Brain-computer interface (BCI) is a new technology aimed at providing efficient means to integrate the human brain and the environment. BCI has been considered as an important approach to help patients with movement difficulties. It has been widely studied for controlling external devices directly by recorded neuronal signals. Here we examine, by neural network simulation, how to apply BCI in order to facilitate trainings during the process of rehabilitation. Specifically, we trained a recurrent neural network to control movement. Then some of the neurons were disabled to mimic the situation of impaired motor cortex. Then two approaches of using BCI to retrain the network for functional recovery were tested. In the first one, BCI was used to interfere with the neurons remained intact in the motor cortex directly. In the second approach, BCI was used to influence the activity of motor cortex through non-invasively manipulating the activities of somatosensory cortex. In both approaches, functional recovery was satisfactory. These results shed new light on the possibility of using BCI to facilitate the rehabilitation of movement control after strokes or other brain injuries.