Studies applying the parallel distributed processing (PDP) model to Japanese kanji (Ijuin, Fushimi and Tatsumi, 2002a) pre-assume that each kanji interactively connects to orthographic, phonological and semantic representations on one-to-one basis. However, Japanese kanji greatly varies their phonological structure. Thus, Experiment 1 conducted the naming task for single kanji Kun-readings of one to three morae, suggested no difference among three mora length. When the same kanji were presented in hiragana, the mora length showed the effect. Likewise, Experimental 2 conducted the same task for single kanji On-readings of one to two morae with high and low kanji-printed frequency. The result showed no effects of mora length, but there were effects of kanji frequency. This result reversed when the same kanji were presented in hiragana: there were effects of more length, but no effects of kanji frequency. Consequently, single kajji orthographical representations correspond to single kanji phonological representations, supporting the application of kanji processing to the PDP model.