Traditionally, it has been proposed that the hippocampus and adjacent medial temporal lobe cortical structures are selectively critical for long-term declarative memory, which entails memory for inter-item and item-context relationships. Whether the hippocampus might also contribute to short-term retention of relational memory representations has remained controversial. In two experiments, we revisit this question by testing memory for relationships among items embedded in scenes using a standard working memory trial structure in which a sample stimulus is followed by a brief delay and the corresponding test stimulus. In each experimental block, eight trials using different exemplars of the same scene were presented. The exemplars contained the same items but with different spatial relationships among them. By repeating the pictures across trials, any potential contributions of item or scene memory to performance were minimized, and relational memory could be assessed more directly than has been done previously. When test displays were presented, participants indicated whether any of the item-location relationships had changed. Then, regardless of their responses (and whether any item did change its location), participants indicated on a forced-choice test, which item might have moved, guessing if necessary. Amnesic patients were impaired on the change detection test, and were frequently unable to specify the change after having reported correctly that a change had taken place. Comparison participants, by contrast, frequently identified the change even when they failed to report the mismatch, an outcome that speaks to the sensitivity of the change specification measure. These results confirm past reports of hippocampal contributions to short-term retention of relational memory representations, and suggest that the role of the hippocampus in memory has more to do with relational memory requirements than the length of a retention interval.