This article describes a study conducted to explore rater training effects of the inexperienced raters on speaking tests, Public English Test System Level 1 (PETS-1), by comparing the quality of scoring before and after training with Many-Facet Rasch Analysis. The analysis shows that, the raters achieve a significant increase in exact agreement percentage between inexperienced raters and expert judges after training. After training, too strict teacher ratings on speaking made appropriate adjustments, all teachers get the good Infit and Outfit values which are within the acceptable range. In general, rater training for PETS speaking tests reveals positive effects and ratings are improved. Many-Facet Rasch analysis helps discover too loose, too strict, and poor fitting raters, as well as abnormal scoring, which all provide a reliable basis for carrying out targeted training to raters.