This study aimed to measure the effect of rater training given to improve the peer assessment skills of secondary school students on rater behaviors using the many-facet Rasch Measurement model. The research employed a single-group pretest-posttest design. Since all raters scored all students, the analyses were carried out in a fully crossed (s x r x c) pattern. There were three facets in the research: student, rater, and criteria. The study group consisted of 25 seventh-grade students at a public school in Ankara in the 2021-2022 academic year. All 25 students in the study group were instructed to write compositions. The compositions were examined by the researchers, and 10 were selected for peer assessment. Before the experiment, students were asked to evaluate their peers' writing skills according to the rubric developed by the researchers. Then, rater training was given to the students for four weeks. After the rater training, the students were instructed to re-evaluate the writing skills of their peers. In the research, four rater behaviors were examined: rater severity, rater leniency, differentiated rater severity, and differentiated rater leniency. When the research results were examined, it was observed that rater training contributed to reducing severity, leniency, and differentiated severity and leniency behaviors.