Problems were presented to 13 university students in which letters were to be added or subtracted (e.g., B + D = ? or F - ? = D). After each problem, each subject gave a retrospective protocol indicating the way in which the problem was solved. Models of performance by each subject in each experimental session shared major properties; choices by all subjects depended on a few features of the problems. Individual differences consisted in the features that were included in a subject's choice procedure and the criteria applied in choices based on the features that were used. (Author)