In this paper, we introduce a spatial appearance group feature for human re-identification in the videos. The observation that motivates this group feature is that the co-traveler can serve as a clue to reduce the ambiguity of human reidentification task. The spatial appearance group feature encodes the group appearances and group structure, and we estimate the probability of two observed groups in different cameras by combining the score predicted by appearance model and group model, which are trained using the observed person and group samples. We conduct experiments to demonstrate the improvements brought by the proposed spatial appearance group feature. We also compare our approach with state-of-the-art group information based re-identification, and the results shows our approach has overall better performance than state-of-the-art.