This article deals with the definition and detection of particular kinds of temporal patterns in behavior, which are sometimes obvious or well known, but other times difficult to detect, either directly or with standard statistical methods. Characteristics of well-known behavior patterns were abstracted and combined in order to define a scale-independent, hierarchical time pattern type, called a T-pattern. A corresponding detection algorithm was developed and implemented in a computer program, called Theme. The proposed pattern typology and detection algorithm are based on the definition and detection of a particular relationship between pairs of events in a time series, called a critical interval relation. The proposed bottom-up, level-by-level (or breadth-first) search algorithm is based on a binary tree of such relations. The algorithm first detects simpler patterns. Then, more complex and complete patterns evolve through the connection of simpler ones, pattern completeness competition, and pattern selection. Interindividual T-patterns in a quarter-hour interaction between two children are presented, showing that complex hidden T-patterns may be found by Theme in such behavioral streams. Finally, implications for studies of complexity, self-organization, and dynamic patterns are discussed.