Bridge systems are the main area of concern of this study since their failure is usually linked to several critical economic and life losses. The bridge structure's failure is defined as the incapability of the bridge's capacity to hold the total applied internal and external loads. Designing bridges to resist different frequent and unexpected load effects has been a concern to multiple agencies. Estimating bridge structures' probabilities of failure is crucial to assess the main causes of bridge failure and thus helps to minimize its probability of occurrence. Subsequently, the main goal of this study is to build a Bayesian network predictive model based on the knowledge gathered from historical bridge failure databases and previous studies. This study explored the suitability of BNs in evaluating the probability of each level of failure for bridge systems and estimating the effect of the different failure causes. Four results validation and assessment stages were implemented, namely, testing the BN through a hypothetical scenario of bridge failure based on a set of evidence, applying it to real-life case studies, executing an in-depth sensitivity analysis to analyze the effect of the occurrence of each of the bridge failure causes on the probability of each of the failure levels, and finally evaluating the maximum a posteriori assignment of each piece of evidence to identify the most likely state estimation of several variables given another different group of evidence in a Bayesian network. The obtained results highlighted the usefulness of the Bayesian network in the study of bridge failure from a probabilistic perspective to estimate the probability of each type of failure effectively and as a tool in forensic engineering and investigations. The developed BN model can be employed by several stakeholders to predict the probability of any potential level of failure based on a set of observed pieces of evidence.