The purpose of this study is to evaluate student’s potential ability on mathematics ability using Many-facet Rasch Model. The test students took consisted of four ranking levels. Three hundred and fourteen elementary students, and three raters were participated in this study. The results show that even with training and delineating a standard for grading, there remained differences in grader severity among the raters. Many-facet Rasch analysis enabled calibration of grader severity for researchers to detect grader irregularities. Data showed that the standard residuals were randomly distributed, indicating that there were no obvious systematic errors. Overall, the study show Many-facet Rasch model can be quite a useful analytic tool in performance assessment.