Abstract: An important problem of using evolutionary algorithm to discover community structure in complex networks is how to reduce the search space of network partitions for speeding up convergence. This paper presents an approach to similarity measurement between nodes and communities based on the local topology information of network nodes, and proposes a new particle swarm optimization algorithm to detect fuzzy communities of network. In the iterative process of algorithm the position vector of particle is modified according to similarity degrees between nodes and communities to promote search efficiency. Experiments on various scale computer-generated networks and real world networks show the capability and efficiency of the method to find the fuzzy community structure of network.