We present a method to automatically detect collaborative patterns of student and tutor dialogue moves. The method identifies significant two-step excitatory transitions between dialogue moves, integrates the transitions into a directed graph representation, and generates and tests data-driven hypotheses from the directed graph. The method was applied to a large corpus of student-tutor dialogue moves from expert tutoring sessions. An examination of the subset of the corpus consisting of tutor lectures revealed collaborative patterns consistent with information-transmission, information-elicitation, off topic-conversation, and student initiated questions. Sequences of dialogue moves within each of these patterns were also identified. Comparisons of the method to other approaches and applications towards the computational modeling of expert human tutors are discussed.