This paper proposes a QoS prediction approach used to recommend web service to solve the problem of choosing high quality web service. This approach bases on other users’ experience. According to the different feeling of different user to the same service, and the different similarity among the QoS of different services, the approach first divides the users and services into several modes, and then computes the means of one mode to predict the QoS of a service that the user never used before. Experiment simulation show that the approach is operable to some extent.