What sort of communication skills should medical and health professionals be taught to develop and assessed on? What informs decisions about the guidance that informs communication training manuals? And does simulation or role-play work for training and assessment? I will address these questions in my talk. I will describe conversation analysis, as a method for analysing real encounters and identifying what works and what is less effective in communication. I will show how conversation analytic research findings underpin a training approach I have developed called the ‘Conversation Analytic Role-play Method’ (CARM). Drawing on a wide range of settings, from dating, police interviews, sales and hostage negotiation – as well as healthcare encounters – I will argue that what we think we know about talk is often wrong, and what works can be found in the tacit expertise displayed in actual interaction between professionals and their clients, patients and service users.