The first version of an “intelligent computer-aided instruction— program built on MYCIN-like expert systems has been implemented. This program, named GUIDON, is a case method tutor in which the problem-solving and tutorial dialogue capabilities are distinct. The expertise to be taught is provided by a rule-based consultation program. The dialogue capabilities constitute teaching expertise for helping a student solve a case. In this paper we describe the rule-based formalism used by MYCIN-like programs, and then argue that these programs are not sufficient in themselves as teaching tools. We have chosen to develop a mixed-initiative tutor that plays an active role in choosing knowledge to present to a student, based on his competence and interests. Furthermore, we argue that is desirable to augment the domain expertise of MYCIN-like programs with other levels of domain knowledge that help explain and organize the domain rules. Finally, we claim that it is desirable to represent teaching expertise explicitly, using a flexible framework that makes it possible to easily modify tutorial strategies and communicate them to other researchers. The design of the GUIDON program is based on natural language studies of discourse in AI. In particular, our framework integrates domain expertise in tutorial dialogues via explicit, modular tutoring rules that are controlled by a communication model. This model is based on consideration of the student's knowledge and interests, as well as the tutor's plans for the case session. This paper discusses interesting examples of tutoring rules for guiding discussion of a topic and responding to a student's hypothesis based on the evidence he has collected.