Collaborative filtering is a key technique in recommender system and applied widely in E-commerce. In reality, due to data sparseness, similarity of users is computed wrongly, which results that really similar users maybe filtered out while false similar users are exploited to produce recommendation. In this paper, two pre-filling methods based on community, respectively simple pre-filling based on community (PFCI) and pre-filling based on community association (PFCII) are presented to overcome the sparsity. If user-item pair is null, its rating is pre-filling by using our method before traditional collaborative filtering is executed. The experiment shows better performance of our methods.