Focusing on the problem of applying and matching resources under large-scale users and computing resources in grid environment, a kind of recommendation-based grid resource matching algorithm is presented. Many existing grid resource matching and scheduling algorithms have to search and compare every grid computing resource node without considering features of grid resources and users’ behaviors, while recommendation system as widely used means in e-commerce could solve all of these two problems well. To utilize recommendation mechanism could pretreat information of users and resources by translating features of grid resources to eigenvectors of items in recommendation system and setting up a satisfaction grade system considering history records with features described in resources applying process that reflect the users’ behaviors through the frequency users computed in resource nodes. Then, the authors improve SVD-based (singular value decomposition) collaborative filtering algorithm that can give users recommendation resource sets by computing the best approximate resource features to users’ behavior features matrix. Especially, the grid resource matching algorithm could mine latent features from given data, efficiently overcome the extreme sparsity of user satisfaction grade data and make use of feedback information from resources scheduling. The problem of matching a mass of resources is solved in a novel way from a new perspective.