This paper talks about a decentralized approach for diagnosis of discrete event systems based on the plant decomposition. A decentralized structure is used to avoid state space explosion found in centralized structure or decentralized structure with composition step. From plant models, all possible faults are identified to construct abnormal behavior models called diagnosers. Originality of this proposition is also to used model-checking to verify diagnosability of the system. The approach is illustrated using an academic benchmark.