This study examined the effects of self-evaluation on student performance in mathematics. Grade 5 and 6 students (N=300) were randomly assigned in intact classes to treatment and control groups. In both conditions a 2-week cooperative learning unit on probability was implemented. In the treatment condition, students received training in self-evaluation for 6 weeks prior to the probability unit. Treatment students became more accurate in their self-appraisals (ES=0.26 on the posttest and 0.35 on retention), an important finding since overestimates of performance reduce students' willingness to seek appropriate help. The treatment had a negligible impact on mathematics achievement, mainly because teachers preferred to teach self- evaluation skills in domains other than math (social skills and writing). The study demonstrated that self-evaluation training clarifies student understanding of curriculum expectations. The findings also weaken the argument for the consequential validity of authentic assessment practices, at least with respect to student achievement. (Contains 5 tables, 1 figure, and 60 references.) (Author/SLD)