Witnesses to complex events often recall nonexistent objects after being exposed to misleading postevent information. The present series of experiments investigated whether descriptions of these "unreal" memories differ from those of memories based on perception. In Experiment 1 subjects viewed a slide sequence depicting a traffic accident. In one condition, the sequence included a slide involving a yield sign. In a second condition, subjects did not see the sign but merely had its existence suggested. Many subjects in both groups later reported seeing the sign, and these subjects provided verbal descriptions. Descriptions that resulted from suggestion were longer and contained more hedges, more reference to cognitive operations, and fewer sensory details. Experiment 2 replicated these findings with a different object. Experiment 3 investigated judges' ability to discriminate the source of the descriptions based on perception and suggestion. Although judges often employed the appropriate criteria, their performance was only slightly above chance. Experiments 4 and 5 revealed that providing judges with clues regarding differences between perceived and suggested memories facilitated discrimination. The results of these experiments indicate that subtle differences exist between perceived and suggested memories, that people have a minimal ability to detect these differences, and that instructions can improve that ability.