The publish/subscribe system is adapted to the dynamic large-scale distributed computing environment well and will be used widely due to its asynchronous, many-to-many and loosely-coupled communication properties. This paper analyzes the state-of-art of publish/subscribe systems. Many existing systems are classified according to the criteria, such as topology structure, event model and subscription model. Then the key techniques, such as matching algorithm, content-based routing algorithm, formal modeling, quality of service, are explained. Typical pub/sub prototypes and products are compared, and their shortcomings and limitations are discussed. The challenge of enhancing the system intelligence by introducing the event semantic and approximate matching, and the trends of supporting the mobile computing and P2P computing environment is forecasted.