This survey of intelligent tutoring systems describes the components of these systems, different teaching scenarios, and the relation of these systems to a theory of instruction. It argues that the underlying pedagogical approach is to make latent knowledge manifest by using different forms of quantitative modeling: (1) simulating physical processes; (2) simulating expert problem solving, including strategies for monitoring and controlling problem solving (metacognition); (3) modeling the plans behind procedural behavior; and (4) forcing articulation of model inconsistencies through the Socratic method of instruction. Proceeding chronologically, examples of intelligent tutoring systems are described in terms of their internal knowledge representations and the evolving pedagogical theory. It also argues that, although these programs are generally only research projects, examples of what they can do make abundantly clear the long term scientific and software-engineering advantages of the new modelling methodology. The text is supplemented by 13 figures, and 31 references are provided. (Author/EW)