This paper presents an automatic grading prototype system developed as recognition engine for Japanese KANJI Dictation Test which aims to certificate the reading and writing ability of KANJI characters. The system is designed to replace the conventional human-grading process. Different from general handwritten character recognition systems which allow to read incorrect (misspelled) characters, the grading system is required to discriminate miswritten characters more strictly from correct ones. This paper introduces the core processing stages of the system. Focused on the relationship between "Error of judgement" and "Rejection", several discrimination methods based on the character classification approach and the likelihood approach are comparatively evaluated. A grading accuracy of error rate less than 0.5% with rejection rate less than 50% is achieved in the performance evaluation test.