This study analyses the risk of severe (≥ 5 fatalities) accidents caused by natural hazards within fossil energy chains (Coal, Oil and Natural Gas). To assess the risk and its uncertainty, a Hierarchical Approximate Bayesian Computation (HABC) is applied on the data collected in the Paul Scherrer Institute's ENergy-related Severe Accident Database (ENSAD). The HABC method presents all accidents as a multilevel system with modules reflecting specific characteristic (e.g. type of natural hazard and country group). To model probabilities for these modules containing very few data, the proposed method samples from the entire system using a complementary numerical technique for posterior sampling, where a stochastic model takes the place of the likelihood one. Once the posterior distributions for the parameters of interest are estimated, risk indicators (e.g., expected fatality rate), including their credibility intervals, are calculated for different combinations: type of natural hazard and country groups (OECD, EU28, non-OECD). The proposed approach provides a unified framework that comprehensively covers accident risks in energy chains, and allows calculating specific risk indicators, including their uncertainties, to be used in informed decision-making under uncertainty.