To assess the impact of various rating scale formats in rating student teachers, a Likert-type scale (Highly agree/slightly agree/neither/slightly disagree/highly disagree) was derived from the graphic scale currently used by the cooperating teachers and university supervisors. They rated the music student teachers on an eight-point scale from superior to average to inferior. Each format's effectiveness in dealing with the problems associated with rating scales was investigated, with the content of the items remaining constant. Data collected enabled the assessment of the contribution of each a leniency error, a halo effect, and a contrast error to ratings on each scale. By treating ratings as the dependent variable in a three way factorial analysis of variance, with rater, ratee, and item as the three independent variables, estimates of the three types of error are obtained in the first order interaction effects. The rating scale employing the graphic format yielded significant first order interactions representing each of the three types of constant error. The rating scale which employed the Likert-type format yielded only one significant first order interaction, that representing the leniency error. It appears that the Likert-type format was more effective in dealing with the halo effect and the contrast error. However, both formats appear to be susceptible to the leniency error. The 15-item grapic scale and the Likert-type derivation (52 items) are appended. (Author/CF)