Aiming at the long response time, inaccurate recommendation and cold-start problems that faced by present recommendation algorithm, this paper, taking movie recommendation system as an example, proposes a collaborative filtering recommendation model based on user's credibility clustering. This model divides recommendation process into offline and online phases. Offline, it uses the result of user's credibility for clustering and then writes the clustered information into a table in database. Online, finds the cluster that target user belongs to and then gives recommendation. As a whole, the model reduces the response time, improves the accuracy of the recommendation rate, and solves the new user's cold-start problem.