Background: Module (community) structure is a common and important property of many types of networks such as social networks and biological networks.Several classes of algorithms have been proposed for module structure detection and identification, including clustering techniques, modularity optimization, and other methods.Among these methods, the modularity optimization method has attracted great attention and much related research has been published.However, the existing modularity optimization method does not perform well in the presence of unbalanced module structures.Methods: In this paper, we first propose a new metric to better characterize the module structure than other metrics in this situation.This metric is based on the average degrees in and between the subnetworks (modules).The model is formulated as an optimization problem.We then develop an algorithm for the module structure identification based on this metric.Results: We apply the proposed method to synthetic network data sets and the gene coexpression networks constructed from the real gene expression data for yeast and the different tissues of a large sample of morbidly obese individuals.The method shows better performance than the existing popular methods in our tests.It also identifies more modules with significantly better enrichment of functionally related genes in the real data.The differences of the identified modules across the different tissues are also discussed.Conclusions: We propose a novel method for module identification in networks, which can capture the important module properties.This method can be applied as a general method to assist in the analysis of biological networks .