This paper describes how cognitive and computational concepts can be applied to build interface and learner models for long-term tutorial interactions in medical radiology. Key human-to-human tutorial dialogue factors that typically occur in the different stages of skill acquisition are captured through an empirical study. The results of the study are detailed and linked to the design of RUI, an Intelligent Tutoring System for multiple domains of radiological expertise. A brief discussion and future research directions offer a comparative view of the method and tools.