<p>When individuals learn through experience, they tend to choose those options they believe will produce the most favourable outcomes. Research by Harris et al. (J Exp Psychol: General 149(10):1855–1877, 2020) indicated that this inclination to consistently engage with the perceived optimal choice can bias the resulting evidence, potentially perpetuating inaccurate beliefs under certain conditions. It remains unclear, however, whether this bias persistence depends on the valence of the learning context. In the present study, participants completed a two-armed bandit task with identical options, where initial evidence was manipulated to induce the belief that one option was better than the other. Experiment 1 was conducted in a rewarding context, where financial gains served as a reinforcer (as opposed to neutral outcomes), while Experiment 2 featured a negative context, with the omission of an expected aversive event as a reinforcer. Behavioral measures showed signs of a lasting bias in the aversive context of Experiment 2, corroborated by a bias in explicit contingency estimates. No such effects were observed in the positive context of Experiment 1. These findings suggest that outcome valence modulates bias persistence in experiential learning. Implications for how negative contexts may promote the maintenance of irrational aversive beliefs are discussed.</p>