Memory is biased. People, for instance, privilege recalling information that confirms preexisting beliefs. This paper examines how selective memory shapes how people search for information. I propose a bandit model, in which different signal are recalled at different rates of probability. Signals which were never received become indistinguishable from signals which were forgotten. This ambiguity creates two opposing interpretations. The absence of signals which were never known is uninformative. However, their absence is informative when selective memory is attributed as their cause. It suggests that disconfirming signals were suppressed, despite their potential benefits. The first interpretation implies under experimentation, whereas the second induces over-experimentation. Thus selective recall leads the individuals to search for information even when they expect with high probability that it won’t be beneficial –in spite of, and precisely because of, their tendency to forget conflicting evidence.