A linear regression model was used to select items from a pool of 700 arithmetic word problems to be used in a computer-assisted mathematics curriculum for elementary school students. The experimental procedure first involved a stepwise linear regression analysis of a student's performance over a set of 25 problems. The probability correct for each of the 700 problems was then predicted and the next 25 problems were selected for the student by matching predicted probability correct with the probability correct assigned to the student at the beginning of the study. The results showed a close match between the predicted and obtained probability correct when the experimental groups were evaluated. (Author/CH)