Models of written word recognition in languages using the Roman alphabet assume that a word's visual form is quickly mapped onto abstract units. This proposal is consistent with the finding that masked priming effects are of similar magnitude from lowercase, uppercase, and alternating-case primes (e.g., beard-BEARD, BEARD-BEARD, and BeArD-BEARD). We examined whether this claim can be readily generalized to the 2 syllabaries of Japanese Kana (Hiragana and Katakana). The specific rationale was that if the visual form of Kana words is lost early in the lexical access process, alternating-script repetition primes should be as effective as same-script repetition primes at activating a target word. Results showed that alternating-script repetition primes were less effective at activating lexical representations of Katakana words than same-script repetition primes--indeed, they were no more effective than partial primes that contained only the Katakana characters from the alternating-script primes. Thus, the idiosyncrasies of each writing system do appear to shape the pathways to lexical access.