Summary: System safety analysis based on fault tree has been widely used for providing assurance to the stringent safety requirement of safety-critical systems. Generating fault trees from models described in AADL, a promising standard language for modeling complicated embedded system, would realize the automation of system safety analysis which is traditionally performed manually. This paper proposes a whole method for constructing fault trees from AADL models, whose main idea is to extract fault information from AADL models by dynamically tracing the possible fault sources of the specified fault objective, store them into a proposed database structure, and then construct fault trees based on the extracted fault information in the database structure. Further, the challenge posed by the common problems of deadlock and fault tree sharing is resolved by one algorithm called Sharing\_Label in our method. We prove the correctness of the whole method theoretically.