Level crossing (LX) safety continues to be one of the most critical issues for railways despite an ever-increasing focus on improving design and application practices. Accidents at European LXs account for about one-third of the entire railway accidents. They result in more than 300 deaths every year in Europe. Due to the non-deterministic causes, the complex operation background and the lack of thorough statistical analysis based on detailed accident/incident data, the risk assessment of LXs remains a challenging task. In the present paper, a general framework of Probabilistic Risk Assessment and Improvement Decision based on Bayesian Belief Networks (PRAID-BBN) is discussed. The developed framework aims to analyze various impacting factors which may cause LX accidents, and quantify the contribution of these factors so as to identify the crucial factors which contribute most to the accidents at LXs, thus allowing for providing improvement measures to reduce risk and costs related to LX accidents. A detailed statistical analysis is firstly carried out based on the accident/incident data. Then, a Bayesian belief network (BBN) risk model is established according to the statistical results. In order to validate the effectiveness of this approach, we apply the PRAID-BBN framework on the basis of the accident/incident data provided by SNCF, the French national railway operator. The main outputs of our study are conducive to efficiently focusing on the effort/budget to make LXs safer.