This paper proposes a new Petri Nets (PNs) knowledge representation scheme to fast estimate the fault section of distribution systems when a fault occurs. Based on the practical guidelines and the heuristic rules obtained by interacting with the dispatchers of distribution systems, a PNs model is first built to represent the related knowledge about the task of substation fault diagnosis. The PNs model built is then transformed into matrix forms, which are relied on to infer the result of fault diagnosis through simple matrix operations. Due to its graphic representation of the heuristic rules and parallel rule-firing manner via matrix operations, the human expertise on fault diagnosis can be advantageously expressed and exploited by means of the PNs knowledge-representing approach. The system is demonstrated on a practical system of Chung-Hsiao substation at Tainan City, Taiwan. Flexibility and effectiveness of the PNs model have been validated for the fault section estimation.