Optimization procedures for a computer-assisted instruction (CAI) system were developed using iterative development and tests of a series of instructional decision models (IDM). The result was a total systems effort in which the instruction was carried on by a dialogue between a computerized tutor and the student. A profile of the student, student characteristics at the time the course began, a map of the structure of the subject matter, and within-the-course history of information exchanges between the student and the CAI program were the factors examined. As factors and decision rules were increased, the IDM's were extended and retested to isolate the most appropriate combination of elements for the next version. The resulting IDM is adaptive to the different needs of individual students. (CH)