Indoor human localization is an important enabling technology for several intelligent applications. One advantage of a passive localization algorithm is that it can estimate human location without requiring the user to carry an electronic device. Because of complex signal radiation in indoor environments, most localization algorithms adopt the fingerprint approach for indoor passive localization. Fingerprints can enable good performance in single-person passive localization. However, when more than two people are in the target area, the system performance may degrade due to the high complexity of the fingerprint-matching task. In this paper, a hierarchical channel state information (CSI)-fingerprint classification system is proposed for passive indoor multiperson localization. In the training phase for coarse classification, fingerprints with similar CSI are first grouped into coarse classes. Then, a coarse classifier is trained for coarse fingerprint matching. Fingerprints belonging to the same coarse class are then entered into a fine classifier for fine fingerprint matching. Experimental results revealed that the proposed approach achieved good accuracy in 93 configurations involving zero to three people. Furthermore, CSI grouping shows that the similarity of CSI depends on the line-of-sight and the number of people.