In the pre-service training program, impromptu speech is one of the most important ability to develop for school principals. To assess the speech ability, most researchers use the total or average of the raw scores as final scores.  However, when there are many students attending the test, raters need to be divided into several groups to provide grading, and the severity of the raters for different groups may impact students´ score. In this study, the scoring criteria, rater´s severity and person´s ability are all considered into the multifaceted Rasch model. The results show that, even for the trained raters, there still exist subjectivity and different level of rater severity. On the other hand, school principals feel most difficult in content, framework, proper usage of words and time control and feel relative easy in pronunciation, prosody, and appropriate manner in the context of impromptu speech. The ignorance of facets and use the raw total or average score as final score may cause bias or unfairness of the score ranking.