Abstract: Aiming at the issue that traditional fuzzy reliability allocation methods may lead to deviations in allocation results due to neglecting units’ posterior failure probabilities, a Bayesian network-based reliability allocation method for meta-actions of CNC machine tools is proposed. Firstly, function–motion–action decomposition is adopted to achieve mapping from the functional layer to the meta-action layer, reflecting the structural characteristics of the machine tool’s motion system. Secondly, a fault tree for meta-actions is constructed for performance failure analysis, and more accurate prior information is obtained by combining expert scoring with the ordered weighted geometric averaging operator. Finally, the fault tree is transformed into a Bayesian network, through reverse reasoning, the posterior failure probability of the meta-actions is obtained and used as the allocation coefficient to achieve the effective update of prior and posterior information. Case study results demonstrate that, compared with traditional methods relying solely on prior information, the proposed approach can more accurately identify key weak links of the machine tool, and the allocation results are highly consistent with actual failure trends, thereby significantly improving the rationality and effectiveness of reliability allocation.