To study the mechanism of motor deficiency, therapy and rehabilitation following stroke, motor cortex network was represented as a directed graph and implemented as a dynamical system. Complexity was used to express the extent to which a system is both functionally segregated and functionally integrated. Motor cortex network was optimized by pattern search to investigate the change of connection under normal, stroke and rehabilitation state. The simulation results show that motor areas have strong connections in normal state. The connections between motor areas decrease greatly, but connections with other areas and self-connections increase in stroke. Stimulation acting on specified areas help to strengthen motor areas connections and weaken other connections to evolve the system to normal state. It indicates the model can reflect the plasticity of motor cortex qualitatively.