The problem of enabling a teaching machine to adapt to the individual characteristics of students in presenting course material has been considered--general outlines of computer algorithms and decision criteria have been formulated. It is assumed that there exists both an ordered set of concepts which comprise the material to be learned and, for each concept, a network of information blocks which includes questions that adequately measure the student's understanding. Given a detailed history of the block sequences studied and answers given by all students who have taken the course, the current student's partial course history, and a predictive model of student behavior, a computer monitor estimates, for each of the information blocks which might be presented next, the probability that this student will correctly answer the block question. Then a criterion such as maximum rate of learning, maximum percentage of correct answers, or some combination of the two serves as a basis for deciding which block will be studied next. A mathematical model based on intuitive ideas about predicting student behavior and a class of models derived from Bayesian statistics have been developed. Simplified applications involving 29 sixth graders and 20 college students have given encouraging, though inconclusive, results. (RM)