Abstract: Existing collaborative filtering algorithms can not promptly reflect the change of users’ interest. For this reason, this paper introduces the human brain’s characteristics of memory and forgetting to personalized recommendation, and proposes a collaborative filtering algorithm based on memory. The effective use of short-term memory reflects users’ recent interest. Long-term memory emphasizes the importance of users’ early interest. At the same time, it combines the short-term memory with the long-term memory and proposes the reconciled memory, which makes the recommender system adaptively track the change of users’ interest. Experimental results show that the proposed algorithm has high quality of precision and rapid convergence rate and that it overcomes the low efficiency of CF, SCF, AUICF algorithms to some extent.