Role identification for network user plays a very important role in function and behaviour network analysis against criminal groups. Therefore, a network key role identification algorithm based on behaviour accumulation is proposed in the paper. Firstly, behaviour attribution of node (user) in network instance is extracted in the algorithm, description and abstraction of user behaviour by vector mode is performed, simultaneously, K mean algorithm is adopted to make clustering analysis for user behaviour; on the basis of conditions described as above, Hybrid Ant-Colony Algorithm (HBACA) is modified to search relationships between variety of grouping sets in the network instance, and identify each user role in the performing process of network instance. Simulation result shows that the algorithm can extract internal relations of communication behaviours between each network member, in order to obtain key role in a specific network instance.