This paper describes three models of learning theories, derives optimal presentation strategies for the models, and compares the effectiveness of the strategies in two computer assisted instruction experiments. The three models described are: 1) the linear, or response-insensitive model, in which the error probability for a given item depends on the number of times it has been presented; 2) the all-or-none model, in which the error probability for a given item depends on the number of times it has been answered correctly; and 3) the random-trial increments (RTI) model, which is a compromise between the linear and the all-or-none model. One of the computer assisted instruction experiments reviewed compared the strategy derived from the linear model with the all-or-none strategy and found that the all-or-none strategy accounted for significantly higher scores on both immediate and delayed posttests. The second experiment compared all three strategies, and found the RTI strategy to be most effective. The paper recommends the formulation of other learning models which would take into account different variables. (RT)