While individualized learning strategies typically provide large amounts of instructional support, they also reply heavily on learner judgement to determine the amount of support required to achieve an objective. Frequently, these strategies result in high achievers selecting too much support and low achievers selecting too little. Interest in this problem led to the development of the Memphis State Regression Model, which systematically selects the amount of instructional support the learner needs. Three adaptive versions of the model were evaluated: (1) quantity of instructional support and incentives; (2) meaningfulness of problem-solving contexts; and (3) density of narrative text. The first study consisted of five treatment groups: individual prescriptions generated by the model, prescriptions based on ability, low or high levels of instructional support, and nonadaptively-varied instructional support. Results indicated that the adaptive group performed significantly better than any of the other treatments. The second model was evaluated via three studies which adapted problem contexts to the learner's interest. Results indicated that the context-specific groups performed significantly better in all three studies. The third study focused on the application of this model in a self-instructional unit covering 10 algebraic rules taught in an introductory college statistics course. Three versions of instruction were developed and administered to students via print or computer presentation--low, high, or learner control of narrative density. Learners in the computer model took more time with both the high- and low-density treatments; subjects in the computer mode of the learner controlled treatment also selected the high-density narrative more often, suggesting that they had less confidence when learning from information presented via a CRT screen. A list of references and a flow chart of the model are provided. (JB)