Teachers in 23 classes in one district were randomly assigned to two frequently used professional development methods. In the skills training treatment expert presenters provided strategies for teaching students how to evaluate their work. Teachers received three training workshops, a handbook of resources (containing case studies, strategies, and student self-evaluation instruments), two half-days of in-school release time, and information about the beliefs of their students about self-evaluation. Teachers were expected to implement the strategies presented. In the action research treatment teachers received the same resources but the in-service sessions emphasized the process used by a previous group of (CLEAR) teachers to improve their use of student self-evaluation. The CLEAR teachers delivered most of the in-service and acted as mentors to the teachers in the action research condition. Pre- and post-test student surveys and interviews indicated that the action research condition made a more positive contribution to student attitudes toward self-evaluation than the skills training condition. The advantage of the action research condition was attributed to (1) sharing control in the in-service provided a better model of sharing control in the classroom, and (2) the handbook examples provided to teachers in both conditions were generated in earlier action research projects, making the information more accessible to teachers in the action research condition. The modest student impact of the treatments was attributable to the neglect of student cognitions about self-evaluation in the classroom, the short duration of the project (8 weeks) and the dilution of effects through rotary timetables.