Fluents are logical descriptions of situations that persist, and composite fluents are statistically significant temporal relationships (nearly identical with those in Allen’s temporal calculus) between fluents. This paper presents an algorithm for learning composite fluents incrementally. The algorithm is tested with a large dataset of mobile robot episodes. The algorithm is given no knowledge of the episodic structure of the dataset (i.e., it learns without supervision) yet discovers fluents that correspond well with episodes. More generally, the algorithm elucidates hidden structure in time series of binary vectors.