In a Web study, using user-ID and password to specify learner is common. However, because of disguise problems etc., it is difficult to specify whether the learner on the terminal side is a person in question. Therefore, it is necessary to specify the learner by some other methods. In such methods, there are facial recognition, writer recognition, and other physical recognition methods such as fingerprint, iris, etc. The writer recognition is used for the handwriting analysis well as a concise procedure. In this research, the writer recognition was studied under the characters limited to hiragana. And the similarity differences by character types were examined. Moreover, the differences between similarity values obtained from own dictionary and others’ dictionaries were examined. As a result, introducing the stability of the character as one of the identification conditions was found to be necessary. And suitable characters for the writer recognition were obtained. From these results, a new writer recognition method was proposed.