Abstract: Introducing some relational conceptions of complex network and community structure, a new community discovery algorithm based on core nodes detecting was proposed. Integrated feature value and gain function are employed in the new algorithm. Comprehensive eigenvalue which is related to the node degree and clustering coefficient is used to detect core nodes in community and the gain function judges when to get the best partition of community structure. The new algorithm is implemented with C++. Experimental results on the classic datasets demonstrate the feasibility and effectiveness of the algorithm. Further, the parameter can be tuned to get more detailed structure of the complex network than traditional algorithm which is very useful in many situations.