Libraries are important to universities, and they have two primary features: readers as well as collections are highly professional. In this study, based on the experimental study with five millions of users' borrowing records, our discussion covers: (1) the necessity of recommender system for university libraries; (2) collaborative filtering (CF) technique is applicable and feasible; (3) user-based CF technique is preferred over item-based; (4) the performance of applying classical used-based collaborative filtering algorithm; (5) the effectiveness of local recommendation and the great saving of computing resource it may bring potentially. Since the data size used in our experiments is the largest one among similar studies, it is believed a valuable reference on this specific direction. [For the full proceedings, see ED562127.]