The pedagogical and theoretical questions addressed in this study relate to the extent to which native Japanese readers with little or no knowledge of Chinese characters recognize Chinese characters that are viewed as abbreviations of the kanji they already know. Three graphic similarity functions (i.e., an orthographically acceptable similarity, a physical similarity, and an extended physical similarity function) were formulated to predict the Japanese learners' target kanji production. Results showed that the learners' performance was poor, with only approximately 30% correct, and that the extended physical similarity function incorporating character frequency and graphic neighbors more accurately accounted for the target production than other functions. Some pedagogical implications and issues for further research are briefly discussed.