Research in the field of computer based instruction (CBI) has led to conflicting evidence on the efficacy of learner control over instruction. Typical implementations of learner control in CBI allow learners to proceed at their own pace, control the sequence of instruction, choose the amount of practice, or decide the level of difficulty attempted. Benefits of learner control have primarily been in the affective domain, with students spending more time on the task when given control over their own instruction. Results in achievement have been mixed, however, when the learner is allowed control over the amount and sequence of instruction. Recent studies allowing the learner to manipulate contextual variables such as text density have demonstrated learner control to be a viable option in CBI. The implication arises that those variables which do not manipulate the amount of content support are those over which the learner should be allowed control. This paper provides a model for classifying the various learner control variables and suggests recommendations for implementing these variables in CBI software and research. (17 references) (Author/EW)