To increase the accuracy of the neighbor screening in collaborative filtering algorithm, an improved system—collaborative filtering with step screening neighbors (SSN-CF)—is proposed in this paper. This algorithm firstly uses an improved Pearson method to compare the similarity between users. After arranging the data in descending order, the uses' characteristic value is calculated. Only those who surpass the threshold value are selected. Then the system gathers the users who graded the priority set to make up the final neighbor set. Finally the users' grades are estimated and recommendation is made. Experiments have shown that the algorithm can effectively get the most similar neighbor set of target uses. Meanwhile, it is tested that accuracy and stability is improved.