Building learning communities is an e®ec- tive solution to conquer feelings of loneliness and to share experiences and resources with one another quickly and e±ciently for learners in an e-learning environment. Ex- isting community models focus only on each learner's sin- gle or primary interest and ignore other interests a learner may hold simultaneously. This paper proposes a multi- interest self-organizing learning community model based on multi-agent technology. Learners with similar interests are automatically grouped into the same community; and, each learner associates with several communities because of his or her multiple interests. This paper also proposes a novel community construction algorithm by calculating knowledge semantic similarity and clustering learner in- terests; and introduces a dynamic community adjustment algorithm by monitoring the learning activities performed by learners. Experiment results in a real e-learning envi- ronment have shown the e®ectiveness and e±ciency of the model and corresponding algorithms.