Many school-improvement efforts include time for teacher collaboration, with the assumption that teachers' collective work supports instructional improvement. However, not all collaboration equally supports learning that would support improvement. As a part of a 5-year study in two urban school districts, we collected video records of more than 100 mathematics teacher workgroup meetings in 16 different middle schools, selected as "best cases" of teacher collaboration. Building off of earlier discursive analyses of teachers' collegial learning, we developed a taxonomy to describe how conversational processes differentially support teachers' professional learning. We used the taxonomy to code our corpus, with each category signaling different learning opportunities. In this article, we present the taxonomy, illustrate the categories, and report the overall dearth of meetings with rich learning opportunities, even in this purposively sampled data set. This taxonomy provides a coding scheme for other researchers, as well as a map for workgroup facilitators aiming to deepen collaborative conversations.