The research literature on performance appraisal is well established but suffers from two inter-related problems. First, rater training has lacked theoretical justification. This void has resulted in a myriad of training programs, with post-hoc explanations used to understand their relative failures and successes. Second, performance appraisal research has focused on rating errors, not the more important issue of rating accuracy. While recent attempts have been made toward understanding accuracy, these attempts again lack theoretical structure. The problem with these shot-gun approaches is that scientific understanding is not advanced in a systematic manner, since there are few attempts to disprove theories. The present study was an attempt to understand previous performance appraisal findings by incorporating theories and research from the person-perception field of psychology. Specifically, the current 3 x 2 x 3 study manipulated theoretically-based cognitive schema rater training (Person/Event/Control) and rating time (immediate/delay) as between subjects factors and rating instrument format (behavior frequency scale/performance evaluation scale) as a within subjects factor. The major hypothesis predicted that the type of cognitive schema employed by raters as they watched videotaped lectures would interact with the instrument format to affect rating accuracy. It was also hypothesized that rating accuracy would be related to one's memory, such that as time and schemas affected memory that there would be corresponding changes in rating accuracy. Results showed that only one of the three component measures of rating accuracy (differential elevation) was affected by an interaction between rater training and rating format. While recognition bias and sensitivity behaved as predicted (prototypical bias exceeded antiprototypical bias, antiprototypical sensitivity exceeded prototypical sensitivity, antiprototypical bias decreased across time), an improvement in rating accuracy was not associated with a reduction in antiprototypical bias. Overall, the present study could not explain some of the previous research findings on rating accuracy. However it represents an important step toward a theoretical integration of rater training with the rating instrument form.