In order to produce relevant responses in conversation, participants monitor turns at talk for the actions they perform, actions such as requests, offers, complaints, etc. (Schegloff, 2007). However, the link between turn construction and action is not straightforward. Turns at talk do not contain discrete ‘illocutionary force indicators’, and what subtle action cues are available, such as interrogative syntax, can be overridden by top-down factors like epistemic status (Heritage, 2012). Given that utterances are often underspecified for action, how is it that participants recognize actions so efficiently, as evidenced by the extraordinarily fast transitions between turns (Sacks, Schegloff, & Jefferson, 1974; Stivers et al., 2009; Levinson, 2013)? As the first step in investigating the cognitive underpinnings of action recognition in conversation, we conducted Event-Related Potential (ERP) experiments using short scripted dialogues in Dutch. The excellent time resolution of ERPs allows us to track listeners’ brain responses as utterances unfold. The critical utterances were assertions (e.g., “I have a credit card”) produced in three sequential environments, affording different ascriptions of the action; as an answer to a question, as an indirect rejection, or as a pre-offer. In each case the assertion is used as a vehicle for some other action, and it is “part of competent membership in the society/culture and being a competent interactant to analyze assertions of this sort for what (else) they may be doing at this moment, at this juncture of the interaction, in this specific sequential context” (Schegloff, 2007, p. 35). We tapped into this competence by exploring the time-course of action recognition, using the following rationale: If comprehension at the action level takes place early in the incoming utterance, enabling quick turn transitions, we should find ERP differences between the actions at the first word or the verb. On the other hand, if action comprehension requires analysis of the complete utterance, ERP effects are expected to predominantly occur at the final word. The results indicate that recipients tune in to the action of an utterance as early as 400 ms after first word onset. However, the time-course of speech act comprehension depends on the specific action. Rejections elicit an ERP effect at the first word and the verb, but not at the final word. We take this to show that when the utterance is a second pair part in an adjacency pair sequence – as was the case in the rejections – recipients seem to recognize the action before the final word, even though the final word is a critical part of the propositional content. The pre-offers, on the other hand, do elicit an ERP component at the end of the utterance, suggesting that analysis of the entire turn is needed to understand the action. These findings indicate that utterance interpretation is sensitive to specific actions and how they are organized in sequences. By bridging conversation analysis and neuropragmatics we have come one step closer to understanding language comprehension in its natural habitat, where action is omnirelevant (Schegloff, 1995).