Heterogeneous beliefs often arise among people who have the same information but different payoff experiences. This paper proposes an explanation in which the experience of realized payoffs distort beliefs: gains lead an agent to underweight negative signals about future outcomes and thus to become overoptimistic, whereas losses do the opposite. I experimentally test the model, and find behaviors consistent with its predictions in a design where experienced payoffs carried no informational value for Bayesian updating. The proposed mechanism is distinct from reinforcement learning, confirmation bias, and attention-based explanations.