Human behavior recognition (HBR) has emerged as a core research area in human-computer interaction. In this article, we give a comprehensive introduction to HBR using Wi-Fi channel state information. We first comprehensively review the state-of-art of HBR, based on the two techniques that drove recent progress in Wi-Fi channel-state-information-based HBR -- fingerprint-based and model-based. Specifically, we describe their corresponding characteristics, general architectures, and provide a performance comparison of the two mechanisms. We then provide insights into the future directions of HBR research, and propose two possible new schemes, and the technical challenges coming with them.