What is the appropriate unit of analysis for the study of classroom discourse? One common analytic strategy employs individual discourse moves, which are coded, counted and used as indicators of the quality of classroom talk. In this article we question this practice, arguing that discourse moves are positioned within sequences that critically shape their meaning and effect. We illustrate this theoretical claim through exploration of a corpus of over 7000 discourse moves in primary literacy lessons. First, we use conventional measures such as the proportion of open and closed questions, and show how these indicators can be misleading when abstracted from the sequences in which they are embedded. We propose a complementary method, lag sequential analysis, which examines how discourse is sequentially structured - i.e. which discourse moves are followed by which other moves, and which chains of moves occur more frequently than expected by chance. We illustrate this method through re-analysis of our corpus of literacy lessons, examining differences between the sequential patterns found in the different classrooms observed. While lag sequential analysis does not resolve all problems inherent in systematic observation of classroom discourse, it does shed light on critical patterns in the data-set that would have otherwise gone unnoticed.