Japanese rendaku 連濁 appears in the second elements (E2s) of compounds, “replacing” a voiceless obstruent with a voiced obstruent (e.g., /tama/ 玉 ‘ball’, /me+dama/ 目玉 ‘eyeball’). Lyman’s Law is ordinarily stated as a phonological constraint preventing rendaku when E2 already contains a voiced obstruent. An alternative proposal is that the constraint is orthographic, preventing a second diacritic in the kana representation of an E2 that already contains one. The crucial cases involve E2s with a medial /p/ — a voiceless obstruent written with a diacritic. We report two versions of an experiment involving compounds with made-up Sino-Japanese binoms 漢 語 二 字 熟 語 as E2s. Speakers chose between forms with and without rendaku. In the audio version of the experiment, E2s with medial /p/ patterned with E2s with medial voiceless obstruents, but in the written version of the experiment, E2s with medial /p/ patterned between those with medial voiceless obstruents and those with medial voiced obstruents. These results lend some credence to the orthographic version Lyman’s Law, but only when orthographic information is visually present.