Drawing on the methods of conversation analysis (Sidnell, 2010; Sidnell and Stivers, 2012) and the data provided by recordings of ordinary interaction, in this paper I ask what a radically empirical approach to word meaning might look like. Specifically, I explore the possibility that we might investigate linguistic meaning through a consideration of interactional troubles. That is, when participants in interaction confront apparent troubles of meaning, what do those troubles consist in? What is the missing something that leaves participants in interaction feeling as though they do not understand what another means? Four types of trouble in interaction are discussed: troubles of exophoric or anaphoric reference, troubles of common ground, troubles of lexical meaning, troubles of sense.