A study was designed to improve fifth-graders' limited use of comprehension monitoring processes during reading through a self-instructional approach. Thirty-nine average and superior comprehenders were tested on their ability to detect inconsistencies contained in short essays prior to, immediately after, and one week after participation in one of three instructional groups: a teacher-directed didactic control group, a neutral self-instruction group, and a specific self-instruction group. Children in the didactic control group received the same instructional content as children in the neutral self-instruction group, but without active rehearsal. In addition to the active rehearsal components, the specific self-instruction group received self-statements specifying an optimal task criterion. Within each ability level, planned contrasts were made among students' immediate and delayed performances across the three instructional groups. Immediately after training, the average comprehenders displayed equivalent improvement across all groups while the superior comprehenders in the specific self-instruction group improved significantly more in detecting errors than did the didactic control group. One week later, no improvement differences were noted between the groups for either ability level. Superior comprehenders displayed significantly greater improvements than the average comprehenders only immediately after receiving the specific self-instruction. Although the superior comprehenders detected more text errors than the average comprehenders, the patterns of improvement following instruction were similar in all the remaining between-ability comparisons. (Author/HOD)