During the retention interval of a working memory task, presenting a retro-cue directs attention to 1 of the items in working memory. Testing the cued item leads to faster and more accurate responses. We contrasted 5 explanations of this benefit: (a) removal of noncued items, (b) strengthening of the cued item, (c) protection from probe interference, (d) protection from degradation, and (e) prioritization during the decision process. Experiment 1 showed that retro-cues reduced the set size effect in a visual recognition task, and did so increasingly with more time available to use the retro-cue. This finding is predicted only by Hypotheses 1 and 2. Hypotheses 3 through 5 were ruled out as explanations of the retro-cue benefit in this experiment. In Experiments 2 and 3, participants encoded 2 sequentially presented memory sets. In half of the trials, 1 item from the first set was retro-cued during the interset interval. Retro-cues improved memory for the second set. This reloading benefit is predicted only by the removal hypothesis: Irrelevant contents are removed from working memory, freeing capacity to encode new contents. Experiment 3 also yielded evidence that strengthening of the cued item might contribute to the retro-cue effect. (PsycINFO Database Record (c) 2014 APA, all rights reserved).