Abstract: In order to accurately predict the accuracy of the compound biaxial turntable, the influencing rule of multiple factors on the system accuracy was analyzed. By using the statistical experimental data of the compound biaxial turntable, the system learns and constructs the Bayesian network structure with the inner axis accuracy, outer axis accuracy, auto-collimation accuracy and the accuracy of the encoder to be tested as the main nodes. The inference model of the system detection accuracy was established in Netica software, and the validity of the Bayesian network (BN) model was verified through evidence sensitivity analysis and mean absolute error (MAE) analysis. The self-learning Bayesian network probabilistic inference is used to analyze the posterior probability changes of each variable of the main target node, and to diagnose and support the reason for the change of the system accuracy. The research results show that the self-learning BN model with compound biaxial turntable accuracy can achieve accurate inference and prediction with system accuracy. And the MAE value of the system accuracy out of tolerance is basically stable within 5%, and the angular interval of 0.125° and the time interval of 20 s are the optimal control parameters of the system, which provides a reference for the application of Bayesian network technology in the accuracy inference of the compound biaxial turntable.