Observers freely viewing a multi-object scene in preparation for a memory test often shift their gaze back to previously inspected objects. These refixations were analyzed to determine if they served a working memory rehearsal function. A gaze-contingent display paradigm was used to limit the observers' viewing of a study scene, which was followed by a spatial probe at one of the object locations and then by a 4AFC test for the probed target. The results indicated a 16% accuracy benefit linked to target refixation that disappeared if 6 or more objects were fixated after the target during study. We interpret these findings as evidence for a monitor-refixate system and a moving-anchor rehearsal strategy. According to this view, observers continuously monitor the availability of items in working memory, then refixate an item when its availability drops below a task-specific criterion. Items are shuffled in and out of this rehearsal set as the rehearsal anchor is systematically moved during scene inspection.