This study applies holistic and analytic scoring methods for elementary school students\'
writing assessment and investigates relative influence of each error source in both methods
which affects writing scores. It shows the inter-rater reliability and the optimal measurement
condition with various tasks, raters, and criteria to improve reliability in writing assessment. 43
elementary school students\' writing papers were analyzed by generalizability theory. The results
of this study are as follows; First, the component which has the largest influence to writing
score was the student\'s ability. Second, the major source of error was the interaction between
person and task in holistic method and the interaction among person, task, and criteria in
analytic method. Third, there was high correlation between holistic and analytic methods and
consequently students were similarly ranked in both scoring methods. The inter-rater reliability
of analytic method was slightly higher than the holistic method, but the difference was not
significantly big. Fourth, increasing the number of tasks or criteria seemed to make
generalizability coefficient higher and measurement error lower than did increasing the number of
raters. The reasonable level of reliability can be gained through well designed scoring rubric, and
well trained raters in elementary writing assessment.