Accident modeling is a vital step, which helps in designing preventive measures to avoid future accidents, and thus, to enhance process safety. Bayesian networks (BN) are widely used in accident modeling due to its capability to represent accident scenarios from their causes to likely consequences. However, to assess likelihood of an accident using the BN, it requires exact basic event probabilities, which are often obtained from expert opinions. Such subjective opinions are often inconsistent and sometimes conflicting and/or incomplete. In this work, evidence theory has been coupled with BN to address inconsistency, conflict and incompleteness in the expert opinions. It combines the acquired knowledge from various subjective sources, thereby rendering accuracy in probability estimation. Another source of uncertainty in BN is model uncertainty. To represent multiple interactions of a cause–effect relationship Noisy-OR and leaky Noisy-AND gates are explored in the study. Conventional logic gates, i.e., OR/AND gates can only provide a linear interaction of cause–effect relationship hence introduces uncertainty in the assessment. The proposed methodology provides an impression how dynamic risk assessment could be conducted when the sufficient information about a process system is unavailable. To illustrate the execution of a proposed methodology, a tank equipped with a basic process control system has been used as an example. A real-life case study has also been used to validate the proposed model and compare its results with those using a deterministic approach.