Discrete event systems (DESs) are event-driven systems, which change their discrete states upon asynchronous occurrence of certain events. This dissertation addresses decentralized/distributed failure diagnosis and supervisory control of DESs.;In a decentralized diagnosis architecture, a local diagnoser performs failure diagnosis completely based on its own observations without communicating with others. A notion of codiagnosability is introduced to capture the property that a system should satisfy such that its failure behaviors are diagnosable by one of the local diagnosers within a bounded delay of their occurrences. Algorithms with polynomial complexity in the size of system/specification models are presented for verifying codiagnosability, computing diagnosis delay bound, synthesizing local diagnosers, and online diagnosis using them. Further diagnosis properties are investigated through the introduction of strong-(co)diagnosability and safe-codiagnosability.;In a distributed diagnosis architecture, local diagnosers exchange their individual observation with each other to perform failure diagnosis collaboratively. Finite automata models are constructed to capture communication delays, and the system/specification/sensing models are augmented with respect to the communication delay models. Via those augmented models, a distributed diagnosis problem is converted to a decentralized diagnosis problem. This allows distributed diagnosis analysis to be performed in same as decentralized diagnosis analysis. Also, in the unbounded delay case decidability of the problem is established.;For supervisory control of DESs, prioritized synchronous composition (PSC) based decentralized control and nondeterministic decentralized control are introduced. A PSC based decision fusion rule is more general than the conventional conjunctive/disjunctive decision fusion rule since it has control-authority besides control-capability. Algorithms are presented for existence and synthesis of PSC based supervisors. Computational complexity of the former is polynomial in the size of both system and specification models, while complexity of the latter is polynomial in the size of systems model, and exponential in the size of specification model. By using nondeterministic supervisors, a weaker condition than the condition of controllability together with co-observability is obtained for decentralized control. Algorithms of polynomial complexity are presented for both existence and synthesis of nondeterministic supervisors in target control and range control problems.