As the fast development of the Internet scale, “data overload” has become one of the most critical problems in computer network analysis. Recommender system has been regarded as the most effective method to solve the problem. But most of existing methods just consider the independent feedback of users without considering the relationship between users, which will inevitably decrease the performance of recommender system. Thus, a friendship prediction algorithm for recommender system was proposed to predict the relationship between different users. Firstly the topological and historical interaction information was taken as the features to judge the existence and relationship type of links. Then the feature combination process based on linear regression algorithm and logistic regression algorithm was implemented. Finally, the experiments based on the real data sets of Epinions and Slashdot were implemented. The experiment results show that our approaches perform very well in link prediction problem.