In this work we asked if risk attitudes influence the way agents learn in a probabilistic environment. For that purpose, 31 male students played a version of a well-known game called the multi-armed bandit (with four levers/buttons). We found that after controlling for cognitive abilities (i.e. Raven’s test), risk seekers in gains preferred to explore in this environment, rather than exploit options, even if one of them was clearly more rewarding. We briefly discuss the reasons and some implications for financial decision theories, in particular, for Bayesian and behavioral proposals.