Historically, training for performance appraisal has focused on the same issues as instrument development--the reduction of psychometric errors in ratings. Efforts were centered around teaching people to use rating scales properly. A review of the literature shows these programs met with mixed success. While a meta-analysis of these data are premature, several hypotheses may be drawn: (1) knowledge of the job in question is more important than rating skills; (2) observational skills are important in real-world ratings; (3) the purpose and context of ratings are as or more important to accuracy than the training itself; and (4) accuracy should be the primary goal of training. Training for performance appraisal is far from universal. Most training efforts in actual use involve learning how to use a particular form or system. One possible training method to improve accurate evaluations involves the use of multiple performance examples, such as videotape, to represent multiple levels of accomplishment. Little systematic knowledge exists about the mechanics of implementing a theoretically-based appraisal system. It is necessary to understand how the appraisal system functions in the operation of the organization. Considerations of equity, of the multidimensionality of job performance, or the cost of more refined observations may make more sophisticated measurement impossible to achieve. Reliable, valid measures that provide accurate determination of two or three levels of performance are an advance over biased assessment of five, or six or more. (JAC)