Despite the prominence of cognitively oriented research in interpersonal communication, relatively little is known about the interplay of cognitive and communicative processes during conversation. For interpersonal communication scholars or at least those interested in interaction, the study of cognition is most useful if it reveals how interactants produce, monitor, modify, and process messages while engaged in conversation. The failure to collect data about in-process conversational cognition is partially due to excessive reliance on what have been identified as “implementation level” cognitive theories and models. Such theories are primarily concerned with specifying cognitive “architectures” and provide little guidance about how cognition is adapted to conversational tasks. However, a dearth of methodological creativity is also partially to blame for the lack of data about in-process conversational cognition. On encountering the admittedly formidable methodological difficulties of collecting data about conversational cognition, researchers have tended to rely on traditional but inadequate methods, or worse, have simply ignored theoretical questions pertinent to cognition in-process. This article suggests that claims about cognition can be made at four levels—biological, implementation, algorithmic, and rational—but that theoretical claims about conversational cognition can most usefully be pursued at the rational and algorithmic levels. In an effort to promote more research at these levels, criteria for studying conversational cognition are proposed, and promising methods are reviewed.