Purpose: In Repeated Search tasks, subjects search repeatedly through the same display. Search remains inefficient even after hundreds of trials. In Memory Search, subjects search through a list of objects in memory. Memory search, initially inefficient, becomes efficient over many trials. In Repeated Search, why do subjects continue to perform an inefficient visual search when they could perform more efficient memory search? To investigate this, we have developed a hybrid visual and memory search task: Panoramic Search. Method: A set of 5 or 8 objects is shown repeatedly in a computer generated realistic scene. The subject's view shifts from one room to another, as if by a head movement. During the shift, different objects become hidden and visible. On each trial, Ss are asked about the presence of an item, presented as a probe picture at fixation. Two versions were run. In the Memory Task, Ss report if they know that the probe object is present somewhere in one of the rooms whether or not it is currently visible. In the Visual Task, Ss respond affirmatively only if they see that the object is currently visible. Results: The slopes of RT x Set Size functions were comparable for Memory and Visual Tasks (about 12 msec/item). Overall, Memory Task RTs are faster than Visual Task RTs. Moreover, in the Memory Task, it did not matter if the target was currently visible or not. The data suggest that subjects relied on memory even for visible items. Conclusion: O'Regan (1992) has argued that observers do not memorize the visual world because the world serves as an "outside memory", available for consultation. Here we have found a condition where a highly reliable memory serves as an internal world. We have persuaded Ss to trust their memory instead of using their eyes.