Feedback on clinical performance of residents is seen as a fundamental element in postgraduate medical education. Although literature on feedback in medical education is abundant, many supervisors struggle with providing this feedback and residents experience feedback as insufficiently constructive. With a detailed analysis of real-world feedback conversations, this study aims to contribute to the current literature by deepening the understanding of how feedback on residents' performance is provided, and to formulate recommendations for improvement of feedback practice. Eight evaluation meetings between program directors and residents were recorded in 2015-2016. These meetings were analyzed using conversation analysis. This is an ethno-methodological approach that uses a data-driven, iterative procedure to uncover interactional patterns that structure naturally occurring, spoken interaction. Feedback in our data took two forms: feedback as a unidirectional activity and feedback as a dialogic activity. The unidirectional feedback activities prevailed over the dialogic activities. The two different formats elicit different types of resident responses and have different implications for the progress of the interaction. Both feedback formats concerned positive as well as negative feedback and both were often mitigated by the participants. Unidirectional feedback and mitigating or downplaying feedback is at odds with the aim of feedback in medical education. Dialogic feedback avoids the pitfall of a program director-dominated conversation and gives residents the opportunity to take ownership of their strengths and weaknesses, which increases chances to change resident behavior. On the basis of linguistic analysis of our real-life data we suggest implications for feedback conversations.