The development of learning theory and its application to computer-assisted instruction (CAI) are described. Among the early theoretical constructs thought to be important are E. L. Thorndike's concept of learning connectisms, Neal Miller's theory of motivation, and B. F. Skinner's theory of operant conditioning. Early devices incorporating those concepts included testing machines and aids developed by Pressey and Peterson in the 1920's and 1930's, and more recently by Skinner. The concept of optimization in learning systems is considered. A distinction is drawn between short-term and long-term optimization--the former having to do with the best procedures for learning a small, discrete item (e.g. how to spell a single word), whereas the latter deals with overall learning strategies. Different approaches to the derivation of optimums are discussed. Finally, it is noted that much of the work to date on CAI has been based upon the thinking of behaviorists independent of researchers in cognitive learning fields. A closer working relationship between these discipline orientations is called for now. (DGC)