The generalizability of scores from different scales in performance assessment was studied. First, a concept map of teachers' and raters' perceptions about various scores and scales was constructed using multidimensional scaling analysis. Then, a generalizability study using a random, partially nested design was conducted to analyze the differences in the various rating systems. This study estimated the variance component of tasks, raters, and evaluative factor based on the scoring systems and determined the optimal number of grading conditions of each facet that maximized the generalizability coefficient. Data for the concept map were from questionnaires completed by about 218 middle school teachers in Korea. Data for the generalizability study were from two different scoring systems used to rate a report and presentation by each student in a middle school social studies class in Korea. The scores of 188 random samples used in the study were the interim scores of each factor before summing up a total score. Results show that the scoring of the performance task using the different rating systems was very consistent from rater to rater. However, the relatively large variance components suggested that the written report was rated differently across the different systems. Findings also suggest that when the student's report or presentation was being assessed, the generalizability of scores was enhanced by combining the ratings from more than one rater, mainly because this effectively increased the number of factors being evaluated. For ratings of performance, the generalizability coefficient increased considerably as the evaluative factors for the scoring standard became more specific. (Contains 19 references.) (SLD)