Coding social interaction has become increasingly attractive for conversation analysts interested in mixed-methods research as a way to demonstrate the robustness of qualitative findings, test relationships between interactional and exogenous variables, and reach a wider audience. However, coding is valuable to conversation analysts only when it is done in a way that attends to participants’ orientations to the phenomenon of study. The puzzle then is how to turn the messy richness of conversational data into codes that are interactionally meaningful and valid. In this article, we draw on the existing literature and our own past projects to discuss opportunities and challenges involved in coding social interaction, with an emphasis on three main aspects of the process: constraining a phenomenon by sequential and formal criteria; transforming behavior into variables; and identifying social actions. Data are in English and Italian.