Abstract: A composite risk analysis model was put forward to analyze the rail haulage accident in inclined tunnel systematically. The model combines preliminary hazard analysis-layer of protection analysis (PHA-LOPA) with bow-tie analysis together based on Bayesian network. First， taking advantage of two-way reasoning ability of Bayesian network identified the risk Bayesian accident nodes with the help of GeNIe software. Second， the accident nodes was studied using PHA-LOPA to determine the causes and results of the accident， and to set independent protection layer to reduce the risk level of the accident nodes. Third， bow-tie analysis was performed on the accident node where the residual risk level was still high， safety barriers were set and the occurrence of the accident was further controlled. The proposed composite risk analysis model was applied to a rail haulage accident in inclined tunnel in a mine， and the validity and feasibility was verified by the results.

Key words: Bayesian network, preliminary hazard analysis, layer of protection analysis, bow-tie analysis, rail haulage accident