With the significant advances in online social networks which provide precious knowledge for personalized recommendation, it is necessary to design effective and efficient method to deal with such data. In this paper, we focus on the recommendation system to integrate the preference information and trust relations among users, following the trust-based recommendation model (TbRM) [1] which considers both the user's reputation and his/her nearest neighbors in the social network. Our main contribution is to present a fast algorithm to solve TbRM model with the aid of graph vertex programming (GVP) in parallel. The idea is to represent the preference and trust information as a graph with both user and item vertices, and its edges contain the preference and trust information. In this case, the recommendation problem becomes a graph analysis procedure which iteratively updates the vertices' states in parallel with the aid of predefined vertex state function and edge information. A series of experiments on four real-world recommendation datasets (Ciao, Epinions, Douban and Flixster) have shown that the graph parallel operations obviously speed up the recommendation procedure, e.g., GVP performs 1.1–2.3× faster than the existing popular distributed stochastic gradient descent algorithm with different number of cores.