Abstract: Radio frequency identification (RFID) provides fast collection of large volume of data and can be used to identify physical objects with unique IDs. In order to provide semantically meaningful data to different applications, RFID data need to be processed to discover user-defined complex events. The Petri-net-based method for the detection of complex events in RFID is proposed. A model named ED-net is introduced to specify semantics of complex events, which is also taken as the basis for the implementation of an event detector. Formal model ED-net is an extension of ordinary Petri net, providing user-defined types, functions and expressions, which are suitable for the precise description of attributes and constraints of RFID complex events, with non-temporal, temporal and parameterized constraints. Taking advantage of the token-flow mechanism of ED-net, we develop a step-by-step detection method that signals the occurrence of a complex event as soon as a corresponding place in ED-net is marked. Through modeling all the events to be detected in one ED-net model, multiple detections of common sub-events of different complex events are avoided. Finally, the experimental evaluations demonstrate the obvious reduction of average complex event detecting latency and CPU resource occupation; The efficiency and advantages of this detection method are verified.