Short-term memory for auditorily and visually presented Chinese characters, varying in frequency and homophonic condition, was studied using an immediate serial-recall task. In experiment 1, where character frequency and homophonic condition were not controlled, auditory presentation produced better recall than visual presentation on the recency as well as the prerecency part of the serial-position function. In experiment 2 and 3, where character frequency and homophonic condition were controlled, auditory presentation produced better recall on the recency part of the serial-position function than visual presentation (a modality effect), and visual presentation produced better recall on the prerecency part of the serial-position function than auditory presentation (a reverse modality effect).