We have proposed a concept of Real-time Human Proxy (RHP), which acquires, symbolizes, transfers and represents human information for avatar-based communication. For applying RHP to avatar-based communication, it is necessary for system developers to enumerate human actions to be acquired, symbolized and represented. In this paper, we propose a new method to extract frequently occurring human motion patterns from human motion acquired by an MCS for supporting system developers’ enumeration. The method employs the idea of motif which is applied to one-dimensional data. We extend the idea to multi-dimensional data, or human motion information. Also we present experimental results.