This study compared obtrusive observers and self-reports to a criterion, unobtrusive observers, to assess how accurately they measured tutor performance. Unobtrusive observers were used as a criterion for comparison because it was assumed that their measures were free of biases. Participants were 55 college-level peer tutors. Course sections were randomly assigned to one of four groups: (1) obtrusive observer and self-reports; (2) self-reports only; (3) obtrusive observer only; and (4) control (no obtrusive observer and no self-reports). All groups contained an unobtrusive observer. Tutors received training in tutoring skills and used those skills in a Personalized System of Instruction classroom to provide feedback to students. Tutors' skills were assessed before and after training. Obtrusive observers accurately assessed tutor performance, but self-reports were only accurate after training occurred. The self-report only group showed the least improvement on a test of written skills. No condition affected the amount of knowledge of course material. Six tables and six figures present study findings. Five appendixes contain the instruments used in the study. (Author)