This study examined the effects of training on peer assessment and comments provided regarding oral presentations in EFL (English as a Foreign Language) classrooms. In Study 1, both the treatment and control groups received instruction on skill aspects, but only the treatment group was given an additional 40-minute training on how to rate performances. The results of the correlation difference analyses show no significant differences between the treatment and control groups, but the three instructors are misfitting, presumably because the frame of reference is set by the majority of student data. In the second study, the treatment groups received long training. Again, there are no significant correlation differences between the treatment and control groups; however, all the instructors are not misfitting, which indicates that the frame of reference in the data improved in Study 2. Analyses of comments reveal that the treatment groups are superior in both quality and quantity of comments. Along with a meta-analytic summary, it is argued that peer assessment is a robust system in which instruction on skill aspects may suffice to achieve a certain level of correlation with the criterion variable (instructor), but training may enhance student comments and reduce misfitting raters. (Contains 3 figures, 5 tables and 2 footnotes.)