Abstract: With the application of collaborative filtering technologies and social network in personalized recommendation system, collaborative recommendation techniques based on social network are now made possible. This paper incorporates key users of social network into the traditional collaborative filtering algorithms to solve cold start problem. Also the influence of key users on recommendation accuracy is verified by experiments. Experimental results show that the key users can improve the accuracy of collaborative filtering algorithm which suggests that the key users can be used to alleviate the impact of cold start problem on the recommendation algorithm.