Research on interactions between teachers and students in music lessons has documented a complex interplay of talk, vocalizations and visual demonstrations. This study employed conversation analysis to explore these multimodal features in order to identify some of the pedagogical practices evident within lesson interaction. Specifically, the aim was to examine how conservatoire students and teachers orientate to both learning and performing within the lesson. Video recordings of 18 one-to-one lessons lasting between 50 mins and 2 hours were analysed. Findings suggest that students demonstrate when they are doing learning versus doing performing through the use of restarts, pauses and apologies in their talk and playing, thus indicating to the teacher which version they would like the teacher to hear for assessment and feedback, and which to ignore. The study highlights how conversation analysis, firstly, enables educators to understand how music performance itself is played out like conversation, mapping key speech devices found in talk onto the multimodal features that are evident in playing. Secondly, how talk and embodied actions inform the study of pedagogical music interaction, demonstrating that, in order to understand and interpret the instruction meaningfully, each component must be studied in context with each other, rather than in isolation.