Aiming at the shortage in the present e-learning systems, how to establish an intelligent e-learning system effectively using data mining is discussed. Associations are extracted from mass of user data in order to provide a comprehensive personalized, customized learning sequence. Associations between Users & Courses, Courses & Courses, Users & Users are mined using data mining, and a multi-dimensional network is built on that basis. Multi-dimensional recommendation is utilized to recommend useful courses to users. Experiments show that multi-dimensional recommendation has a high precision and provides a good experience to users.