The theory of expected utility maximization (EUM) proposed by Bernoulli explains risk aversion as a consequence of diminishing marginal utility of wealth. However, observed choices between risky lotteries are difficult to reconcile with EUM: for example, in the laboratory, subjects’ responses on individual trials involve a random element, and cannot be predicted purely from the terms offered; and subjects often appear to be too risk averse with regard to small gambles (while still accepting sufficiently favorable large gambles) to be consistent with any utility-of-wealth function. We propose a unified explanation for both anomalies, similar to the explanation given for related phenomena in the case of perceptual judgments: they result from judgments based on imprecise (and noisy) mental representation of the decision situation. In this model, risk aversion is predicted without any need for a nonlinear utility-of-wealth function, and instead results from a sort of perceptual biasbut one that represents an optimal Bayesian decision, given the limitations of the mental representation of the situation. We propose a specific quantitative model of the mental representation of a simple lottery choice problem, based on other evidence regarding numerical cognition, and test its ability to explain the choice frequencies that we observe in a laboratory experiment.