Evidence from brain-damaged patients suggests that there are separate buffers for input and output phonological retention in verbal short-term memory (STM). This possible distinction was investigated with college students (Experiment 1 to 3) and deaf signers of American Sign Language (ASL) (Experiment 4) using different verbal materials in a serial probed recall paradigm. It is reasoned that natural linguistic input (speech for hearing people and ASL for deaf people) would be stored in an input phonological buffer whereas internally generated phonology derived from reading, naming pictured objects, or lip-reading would be stored in an output phonological buffer. In this study, participants were presented with memory lists in which presentation modality (spoken vs. lip-read word, written vs. lip-read word, etc.) was changed after every second item. A probe item from the list was repeated at the end of the list and participants were instructed to either recall the item in the list that has immediately followed the probe or recall the first item after the probe that is in the same modality. Some of these same-modality items were temporally distant, that is, having two intervening items of a different modality. It is predicted that the temporally distant probe in the same modality with the target results in higher memory performance than the temporally adjacent probe in a different modality only if the switch in modalities is between input and output phonological forms. The results from Experiment 1 demonstrated that spoken words and written words were stored in the input and output phonological buffers, respectively. The results from Experiment 2 and 3 further supported the hypothesis in showing that written words were retained in the same buffer with lip-read words and with nameable pictures, while spoken words were retained in a different buffer from these materials. The findings from lists consisting of words in ASL and nameable pictures in Experiment 4 were not conclusive. However, preliminary data suggested that there might also be a separation between signed words and nameable pictures. Overall, the findings from this study conformed to the predictions from the hypothesis of separate input and output phonological retention.