The midstream order deficit (MOD) is a phenomenon that accuracy in recalling the relative order in a cycling visual sequence is much lower than when the sequence is presented just once (Holcombe, Kanwisher, & Treisman, 2001). The occurrence of MOD interested in how the relative order is encoded. Chiba & Yokosawa (2003) reported that phonological and visual codes played different roles in order encoding. In this study, to prevent phonological encoding of relative order, four characters with the same pronunciation were used within each target set. By this manipulation, we were able to examine whether MOD would need for phonological encoding. Japanese Kanji characters were used as stimuli, because they have many homophones. Participants were required to report the relative order by connecting the printed characters on the response sheet using arrows. In both experiments, MOD disappeared when the Kanji characters had the same pronunciation, suggesting a critical role for phonological encoding in MOD. In experiment 1, when the stimulus sequence was presented after memorization of the four Kanji characters, the accuracy of the recall of order in both the single and cycling conditions was higher. However there was not the MOD effect. In experiment 2, it is also examined whether MOD would occur when performance improved. It takes longer processing time for a complex character like a Kanji character than for a simple character. A slow final item duration was used in experiment 2. However MOD still was not found. The results contradict the explanation that the disappearance of MOD was related to task difficulty. It is summarized that MOD disappeared when the Kanji characters had the same pronunciation, suggesting an important role for phonological encoding in MOD. The robust MOD effects found with the verbal reports in Chiba & Yokosawa (2003) support this suggestion.