The traditional collaborative filtering recommendation algorithms face the dilemma of severe data sparsity and real time of recommendation, their recommendation quality is not obviously high. To improve recommendation efficiency, firstly, user rating items and similarity measurement method based on cloud model were researched. Then the definition of recommendation system trust constraint based on cloud model was given, and improved the constraint function of subjective trust cloud model and trust change function of trust change cloud model. Finally, a collaborative filtering recommendation algorithm based on cloud model was put forword. The experimental results show that the algorithm still obtains good recommendation efficiency on situation of user rating data sparsity compared to the traditional algorithms, it has high utility.