AbstractAlthough the influence of psychosocial hazards on mental health has been widely recognized, the complex pattern of psychosocial hazards in the prediction of mental health still needs to be identified. Bayesian networks serve as an important, and as yet untapped, method to achieve this aim. As such, a Bayesian network (BN) was developed to assess the influence of psychosocial hazards on mental health based on survey data from 186 site-based construction practitioners and informed by the literature and expert knowledge. Results indicated that 66% of this target population suffered from poor mental health (i.e., mild, moderate, and severe states). Poor physical environment, contract pressure, and lack of coworker support were the root nodes in the model, which played an important role in determining the probability of other psychosocial hazards. Lack of job control, lack of coworker support, role ambiguity, role conflict, and lack of supervisor support were, relatively, the most influential hazards to mental health. Jointly controlling multiple psychosocial hazards was found to be better for increasing the probability of good mental health than controlling one individual hazard. The predictive inference revealed that managing the combination of lack of job control, lack of coworker support, role conflict, and job insecurity achieved the highest probability (77%) of good mental health among all four-hazard scenarios. Based on this particular scenario, further optimizing poor physical environment resulted in the highest probability (83%) of good mental health for five-hazard scenarios. The proposed BN model can be used as a powerful decision support tool to improve mental health in the construction industry via managing a more targeted subset of psychosocial hazards.