Proposed and evaluated in this research were causal models that included measures of cognitive ability, job knowledge, task proficiency, two temperament constructs (achievement and dependability), awards, problem behavior, and supervisory ratings. The models were tested on a sample of 4,362 U.S. Army enlisted personnel in nine different jobs. Results of LISREL analyses showed partial confirmation of Hunter's (1983) earlier model, which included cognitive ability, job knowledge, task proficiency, and ratings. In an expanded model of supervisory ratings, including the other variables mentioned, technical proficiency and ratee problem behavior had substantial direct effects on supervisory ratings. Ratee ability, job knowledge, and dependability played strong indirect roles in this rating model. The expanded model accounted for more than twice the variance in ratings in the present research than did Hunter's variables alone.Measures of criterion performance are necessary for almost all personnel research applications in organizations. Criteria are needed to assess the effectiveness of personnel selection procedures, organizational training programs, job design efforts, and many other personnel-related actions and interventions. Furthermore, performance rating scales are by far the most often employed criteria for measuring an individual's job performance. Thus, strong motivation exists for obtaining the most accurate ratings possible to use as performance criteria.Historically, attempts to improve performance ratings have included (a) designing rating formats that help raters to increase the precision of their performance evaluations (e.g., Smith & Kendall, 1963) and (b) developing training programs to reduce rating errors and enhance accuracy (e.g.,