In general, systems are formed by the composition of several modules, local components or subsystems, and may exhibit a large number of states. The growth of the global system model with the number of system components leads to high computational costs for diagnosis techniques based on the global model. In order to circumvent this problem, several works in the literature address the problem of modular fault diagnosis of discrete event systems. In these works, the computation of the global system model is avoided, and the diagnosability of the language of the system is inferred from the behavior of its components. In this paper, a new approach to online diagnosis of modular systems is presented. The method is based on the computation of a Petri net diagnoser, called as synchronized Petri net diagnoser (SPND), that is constructed from the nonfaulty behavior of the modules of the system. We also introduce the definition of synchronous diagnosability of the language of a modular system with respect to the languages of its modules, and present an algorithm to verify this property.