Visual evoked potentials (VEPs) are electrical responses to visual stimuli aroused in the occipital region of the cerebral cortex. The P100 is an important component of VEP, and its latency is generally used for diagnosis of many clinical diseases. The existing methods to extract VEP have several drawbacks such as involving too many stimuli, costing too much time or requiring a large training database. However, the realtime applications such as brain-computer interfacing (BCI) require that the VEP should be quickly extracted from fewer stimuli, which is seldom met by those existing methods. In this paper, a VEP extraction method involving only a few visual stimuli is proposed based on improved Wiener filter. Through experiments on multiple subjects, it is verified that the proposed method can extract VEP quickly and give a reliable estimation of latency of P100.