The effect of rating scale format (behaviorally anchored rating scales (BARS) and Likert) and rater training on leniency and halo in student ratings of instruction was investigated. The subjects (N=269) were students enrolled in required courses at a graduate theological seminary in the southwestern United States. A repeated measures design controlling for teacher and course was used. Findings indicated that: (1) training was effective in reducing leniency and halo in ratings from both instruments; (2) trained raters exhibited less leniency on two rating dimensions when using BARS than when using the Likert scale; and (3) trained raters exhibited less halo when using the Likert than when using the BARS. The findings demonstrate the importance of focusing efforts to improve quality of ratings on the students rather than on the format of the instrument. Contains 24 references. (Author/KM)