A central issue in recent psycholinguistic research is whether phonological planning units in word production are language-specific. Evidence from some Indo-European language speakers contrasts with recent studies of native Chinese and Japanese speakers, who did not exploit word-onset phonemes in the word-form preparation paradigm. In the present study, we first replicated our earlier finding that no preparation effect for word-onset phonemes occurred when target responses were learned from commonly used Japanese orthographies, including kanji and kana. We then asked participants to learn target responses presented in romanized Japanese, known as romaji, and found a significant phoneme preparation effect. However, we failed to find a similar effect for auditory presentations of the target responses during the learning phase. These results suggest that Japanese speakers, who do not essentially base their speech planning on phonemes, can prepare phonemic segments in special circumstances in which the orthographic representation of romanized Japanese is first activated.