The rapid development of electronic information technology has spawned many complex systems which include a considerable number of components. Fault diagnosis aims to accurately detect faults in these systems in a timely manner, thus providing important support for the rapid recovery of system functions. This paper proposes a decentralized fault diagnosis approach based on Petri nets. Two algorithms for each diagnosis site and the coordinator in the decentralized architecture are provided. In addition, a diagnosis protocol used between diagnosis sites and the coordinator is developed. The diagnosis algorithm of each site is constructed based on Petri net basis markings and vectors. Thus, exhaustive enumeration of all transition sequences consistent with the observed label sequence is avoided, leading to better computational efficiency. The proposed decentralized diagnosis approach enjoys higher computational efficiency and better stability than the traditional centralized methods. Compared with existing approaches, that described here requires fewer assumptions on the structure of a net system and has broader applicability. This work is supported by the National Natural Science Foundation of China (No. 62103349).