To compare reliability and systematic error of estimation of Liket rating scale and Behaviorally anchored rating scales for teachers' teaching assessment. Situations of this research were teachers' ratings and students' ratings of teachers' teaching in steps before, between and after teaching. The groups of samples consisted of 52 mathematics teachers teaching in Mathayomsuksa five and 52 classes of students selected by the teachers. The tools used in this research were the teachers' teaching rating scales as Likert rating scale and as Behaviorally anchored rating scales. The questionnaires were sent to the subjects by mail. To detect reliability, intraclass correlation coefficient was performed and tested the significance by the Wilcoxon signed ranks test. To detect leniency error, the Wilcoxon signed ranks test was used to compare means of rating scores. For halo error, the dependent t-test was conducted. Results of the study could be summarized as followed: 1. The result of the test of reliability revealed that there was no evidence to confirm that behaviorally anchored rating scales had higher interrater reliability than the Likert rating scale. 2. It was concluded that there was no evidence to confirm that behaviorally anchored rating scales had leniency error less than Likert rating scale. 3. It was found that behaviorally anchored rating scales had Halo error less than Likert rating scale in both situations, teachers' ratings and students' ratings, at .05 significance level.
