For some time I have been involved in efforts to develop computer-controlled systems for instruction. One such effort has been a computer-assistedinstruction (CAI) program for teaching reading in the primary grades (Atkinson, 1974) and another for teaching computer science at the college level (Atkinson, in press). The goal has been to use psychological theory to devise optimal instructional procedures—procedures that make moment-by-moment decisions based on the student's unique response history. To help guide some of the theoretical aspects of this work, research has also been done on the restricted but well-defined problem of optimizing the teaching of a foreign language vocabulary. This is an area in which mathematical models provide an accurate description of learning, and these models can be used in conjunction with the methods of control theory to develop precise algorithms for sequencing instruction among vocabulary items. Some of this work has been published, and those who have read about it know that the optimization schemes are quite effective—far more effective than procedures that permit the learner to make his own instructional decisions (Atkinson, 1972a, 1972b; Atkinson & Paulson, 1972). In conducting these vocabulary learning experiments, I have been struck by the incredible variability in learning rates across subjects. Even Stanford University students, who are a fairly select sample, display impressively large betweensubject differences. These differences may reflect differences in fundamental abilities, but it is easy to demonstrate that they also depend on the strategies that subjects bring to bear on the task. Good learners can introspect with ease about a "bag of tricks" for learning vocabulary items, whereas poor