Disclosed is an accurate control method of visual stimuli for a brain-computer interface. It is a common approach for brain-computer interfaces to evoke specific EEG signal patterns by visual stimuli and recognize the EEG signal patterns in real time. However, due to the influence of process scheduling, a process showing the visual stimuli may sometimes be dispatched out of a CPU, leading to the difficulty in guaranteeing the accuracy of the visual stimuli and the recognition effect of the EEG signal patterns. The invention designs a control method to support accurate visual stimuli of a brain-computer interface. A software system implementing the method comprises a generator, an actuator and a controller. The generator automatically generates an image sequence according to test requirements. The actuator is a module running on a GPU. At the beginning of a trail, the controller asynchronously calls an interface of the actuator to start the actuator, and the actuator accurately shows the image sequence generated by the generator. At the end of the trail, the controller asynchronously calls an interface of the actuator to stop showing visual stimuli.