基金广东省自然科学基金重点项目（05100302）With the booming of E-commerce, recommender systems are more and more widely used in this area. Collaborative filtering is one of the major technologies used in recommender systems. Scalability and quality are two major challenges in collaborative filtering recommender systems. This paper addresses the quality issue, which is mostly invoked by the sparsity of datasets. BP neural networks have powerful learning and modeling capabilities. They are effective in processing non-complete information. Borrowing these capabilities from BP neural networks, this study fills in the null values with reasonable predicts, thus decreasing the sparsity of datasets and increasing the recommendation quality of collaborative filtering recommender systems, This work is supported by the Guangdong Natural Science Foundation（No. 05100302）.