Petrol stations are normally located close to residential areas where a large population is present. Therefore, explosion accidents at service stations may lead to significant human losses. This study presents an application of a Bayesian-network-based (BN) quantitative risk analysis (QRA) method to modelling gas explosion process from an initial release to consequent explosions and human losses as the BN is able to reveal complicated mechanisms with complex interrelationships between parameters. The proposed BN of explosion events at petrol stations contains 14 nodes and 18 links, which consider several critical risk factors, including spill, ignition, explosion, evacuation and human loss. Meanwhile, to reduce uncertainties caused by data shortage and improve the reliability and accuracy of the proposed method, three kinds of data practical information, computational simulations and subjective judgements are involved in the quantification of the proposed BN. A case study is provided in this paper, and sensitivity studies are also conducted. The result indicates that human loss can be influenced significantly by release scenarios, ignition sources and time of day. It also proves that the risk-based method is able to provide an effective and reliable assessment of human safety when explosion events occur at petrol stations.