Speech production research has shown that Japanese monolingual speakers use mora-sized phonological units, not phoneme-sized units, when phonologically encoding Japanese words. Recent bilingual research has indicated that proficient Japanese-English bilinguals nevertheless use phoneme-sized units when phonologically encoding English words, suggesting that use of a phonological unit that is smaller than that of their L1 develops with increasing proficiency in English. The purpose of the present research was to determine whether proficient Japanese-English bilinguals also begin to use the smaller, phoneme-sized units when producing Japanese words. In a masked priming naming task, proficient Japanese-English bilinguals produced a significant masked onset priming effect for English words, confirming that they do use phoneme-sized units when phonologically encoding in English (L2). These bilinguals, however, showed only mora-based facilitation for Japanese words in an experiment involving only Japanese words. These results suggest that proficient bilinguals use different unit sizes depending on the language being produced, and that for bilinguals whose L1 and L2 have different unit sizes, the phonological encoding process is at least somewhat different in their two languages.Key words: masked onset and mora priming effect, Japanese-English bilinguals, phonological unit-size, speech production, phonological encoding.When one speaks, representations of to-bespoken words are first accessed in the speaker's mental lexicon, and then the phonological properties of the words are retrieved and encoded before the articulation system initiates the actual act of speaking. According to Levelt, Roelofs, and Meyer (1999), the phonological encoding process occurs incrementally from word beginning to word ending by assigning phonemes to metrical frames, in which the syllable and stress patterns are specified. The information is then sent to the articulatory system for execution of motor movements. The assignment of phonemes into a metrical frame is called the segment-to-frame association process, and is the essential step for successful word production.The first empirical evidence that the phoneme is the unit used in the phonological encoding process was presented by Meyer (1991) using an "implicit priming" paradigm with Dutch stimuli. In these experiments, participants first learned a small set of semantically related word pairs (e.g., hour-time, swim-pool,