Previous studies have shown that the efficiency of visual search does not improve when participants search through the same unchanging display for hundreds of trials ("Repeated Search"), even though participants have a clear memory of the search display. Why don't participants use memory? In earlier experiments, using a search for letters, search efficiency actually improved when we removed the visible stimuli, forcing participants to rely on memory. However, that manipulation might have changed the task. In repeated visual search, participants probably visually located the target on each target-present trial ("The K is right there"). For the repeated memory search, participants might have simply learned a 2AFC mapping of "target" letters to one response key and "non-target" letters to another key ("The K is a member of the target set", "The Z is not a member of the target set"). In the present work, we asked participants to use the computer mouse to click on the location of targets in repeated search for visible or remembered displays. Under these conditions, as before, visual search remained inefficient (39 msec/item) over hundreds of searches through the same, unchanging display. Interestingly, memory search was even more inefficient (81 msec/item). Like visual search, it failed to become more efficient over hundreds of repetitions. It seems clear that participants used visual search because it was a more efficient strategy. Can participants ever use memory? Yes, when targets were restricted to a subset of the visible letters, participants rapidly learned to use memory for that fact to restrict search to the relevant subset.