The hypothesis that task-specific criterion-referenced self-assessment can have a positive effect on students' metacognitive engagement and learning was tested. Seventh graders (n=47) were asked to invent, apply, and explain a classification system for a group of animals. Treatment subjects periodically assessed their performance in terms of a written rubric that listed the criteria for each task and gradations of quality for each criterion. Students in the control group were not asked to assess their work. Think-aloud protocols were collected and coded to provide insight into spontaneous self-assessment, the classification of self-assessment, and the influence of self-assessment on metacognitive engagement and learning. Approximately three-quarters of the students assessed themselves spontaneously. Girls in the treatment group were more metacognitive than boys, but no statistically significant differences were found for boys in treatment and control groups. Treatment students tended to outperform the control group on posttests. The rubric appeared to have a positive effect on the criteria that students used in their spontaneous self-assessments, and students who assessed their own work were remarkably willing to revise it. An appendix contains the scoring rubric given to the experimental group. (Contains 18 tables and 109 references.) (Author/SLD)