Low-probability, high-impact events are difficult to manage. Firms may underinvest in risk assessments for low-probability, high-impact events because it is not easy to link the direct and indirect benefits of doing so. Scholarly research on the effectiveness of programs aimed at reducing such events faces the same challenge. In this article, we draw on comprehensive industry-wide data from the U.S. nuclear power industry to explore the impact of conducting probabilistic risk assessment (PRA) on preventing safety-related disruptions. We examine this using data from over 25,000 monthly event reports across 101 U.S. nuclear reactors from 1985 to 1998. Using Poisson fixed effects models with time trends, we find that the number of safety-related disruptions reduced between 8% and 27% per month in periods after operators submitted their PRA in response to the Nuclear Regulatory Commission's Generic Letter 88-20, which required all operators to conduct a PRA. One possible mechanism for this is that the adoption of PRA may have increased learning rates, lowering the rate of recurring events by 42%. We find that operators that completed their PRA before Generic Letter 88-20 continued to experience safety improvements during 1990-1995. This suggests that revisiting PRA or conducting it again can be beneficial. Our results suggest that even in a highly safety-conscious industry as nuclear utilities, a more formal approach to quantifying risk has its benefits.