How do people form beliefs about novel risks, with which they have little or no direct experience? We address this question using a 2020 representative US survey of beliefs about the lethality of Covid. The survey reveals a number of surprising findings, including most dramatically the evidence that the elderly are much more optimistic about Covid death risks than the young. In fact, the elderly underestimate these risks, while the young hugely overestimate them. To shed light on this and other findings, we present a model in which people selectively and automatically recall past experiences, including experiences from other domains, and use them to simulate the novel risk. This mechanism entails a tradeoff: better recall of an experience enhances risk perception by providing more material for simulation, but dampens risk perception by interfering with recall of other experiences that are perhaps better at simulating the novel risk. The model connects average overestimation of unlikely risks with strong disagreement. It predicts that people exposed to many interfering experiences should underestimate the novel risk and be less sensitive to direct experiences with it. We find empirical support for these and other predictions of our model using our measurement of respondents’ Covid as well as non-Covid past experiences.