With the popularity of unmanned aerial vehicles (UAVs) in different application scenarios, a series of potential risks and challenges have emerged. Few quantitative studies in the existing literature analyze the risk factors that affect and cause UAV accidents. Therefore, this study aims to reveal the dependency relationships among the risk factors associated with UAV accidents by using a data-driven tree-augmented naïve Bayesian network (TAN-BN) method. Firstly, UAV accident reports from transportation departments of various countries are collected and analyzed to establish a dataset on the risk factors inducing different UAV accidents. Then, a data-driven UAV risk analysis model is constructed through data learning and training approaches to reveal the interdependency of risk factors and their comprehensive impacts on the severity of UAV accidents. Sensitivity analysis is comprehensively conducted to show the rationality of the proposed TAN-BN. The results reveal that the risk factors “other aircraft”, “observer errors”, “non-powerplant”, “preconditions for unsafe acts”, “powerplant”, and “adverse weather” are the key factors resulting in UAV accidents. In addition, “observer errors”, “other aircraft”, “non-powerplant” and “adverse weather” are more likely to incur severe “accidents”. Some potential proactive plans and measures are proposed to help prevent the occurrence of UAV accidents.