The safety of modern safety-critical systems is increasingly receiving attention. AADL, as an effective modeling language, is widely used for architectural modeling of embedded safety-critical systems. Currently, the main challenges facing the safety analysis of AADL models are the system’s dynamic behavior, state space explosion, rare event prediction, and the lack of explanation of unsatisfied specifications. To address these issues, we propose QuanSafe, a discrete-time Bayesian network (DTBN)-based framework of quantitative safety analysis for AADL models. The dynamic behaviors and temporal features of AADL models can be described entirely using DTBN. Moreover, DTBN can effectively avoid state space explosion and poor performance when dealing with rare events. At the same time, DTBN has the ability of diagnostic analyses, which helps improve the system. QuanSafe provides a complete algorithm to transform AADL models into DTBN models. In addition, it supports multiple automated safety analysis methods with improved metrics. We conduct a case study on the Aircraft System. The experimental results show that our approach has higher efficiency and more comprehensive analysis capabilities than existing research.