At the beginning of a Java computer programming course, nine students in an undergraduate class and nine students in a graduate class completed a web-based programmed instruction tutoring system that taught a simple computer program. All students exited the tutor with an identical level of skill, at least as determined by the tutor's required terminal performance, which involved writing the program and passing multiple-choice tests on the program's elements. Before entering and after exiting the tutor, students completed a test of rule-based performance that required applications of general programming principles to solve novel problems. In both classes, the number of correct rule answers observed before entering the tutor did not predict the number of learn units that students subsequently used to complete the tutor. However, the frequency of learn units was inversely related to post-tutor rule-test performance, i.e., as the number of learn units used in the tutor increased over students, the number of correct answers on the post-tutor rule test decreased. Since time to complete the tutor was unrelated to learn unit frequency, these data suggest that high achieving students may have generated autoclitic learn units while using the tutor. Interteaching, as an occasion for generating and sharing interlocking learn units, may be an effective complement to programmed instruction in promoting optimal learning in all students. (Contains 2 tables and 4 figures.)