This study examines the rationality of safety beliefs and the correlation between risk perception and actual level of job-related physical risk using data from the Maine’s American lobster fishery. To avoid measurement errors, I applied the instrumental variables to estimate risk perception and the compensation for the captain. Using educational achievement as a proxy for cognitive ability, I find no evidence supporting a correlation between risk perception and cognitive ability. While captains incorporate information on actual commercial fishing accidents to form their probabilistic judgments, I find evidence that the learning process is non-linear in lobster fishery and does not follow an expected Bayesian linear framework. Also, the results suggest that the adjustment process slows down as the new information arrives. I could not confirm that fishermen use normalization strategy to minimize risk perception as a psychological method of coping with the threat.