Concerning the fault diagnosis problem of a discrete event system which is modeled by a labeled Petri net, a structure based approach that utilizes the properties of paths is presented. Assuming the structure of the labeled Petri net is known, faults which are hidden in sensors or executing devices are modeled by unobserved transitions and some other observed transitions that are labeled the same with the regular ones. Firstly, several transformation rules are provided to simplify a priori, the structure of the labeled Petri net, before the diagnostic procedure. Then, multiple path constrains can be derived through partitioning the transformed Petri net into particular paths. Finally, with the corresponding path constrains, some constrain rules can be formulated, then the results of fault diagnosis are characterized after observing a labeled sequence with the use of the constrain rules, moreover, the probability of fireable of the fault for the uncertain situation can be analyzed. One of the contributions of this work is scaling down the memory cost with the transformation rules. The other one is no more seeking the system state space for fault transitions benefiting from the structural constrains. Moreover, the computational complexity of the diagnostic process can be much reduced owing to the limited number of paths.