For decades, researchers have sought to understand the organizing principles of auditory and visual short-term memory (STM). Previous work in audition has suggested that there are independent memory stores for different sound features, but the nature of the representations retained within these stores is currently unclear. Do they retain perceptual features, or do they instead retain representations of the sound's specific acoustic properties? In the present study we addressed this question by measuring listeners' abilities to keep one of three acoustic properties (interaural time difference [ITD], interaural level difference [ILD], or frequency) in memory when the target sound was followed by interfering sounds that varied randomly in one of the same properties. Critically, ITD and ILD evoked the same percept (spatial location), despite being acoustically different and having different physiological correlates, whereas frequency evoked a different percept (pitch). The results showed that listeners found it difficult to remember the percept of spatial location when the interfering tones varied either in ITD or ILD, but not when they varied in frequency. The study demonstrates that percepts are the units of auditory STM, and provides testable predictions for future neuroscientific work on both auditory and visual STM.