The use of complex networks, as efficient models, for describing a wide variety of natural systems and phenomena is increasingly extending. For investigating a network, perceiving its structure is extremely important and is also necessary for understanding its function. Communities are of the most significant structural concepts. I n communities there are plenty of (and often strong) relations between the members of the same community, in contrast to sparse connection between members of different modules . This causes that the communities show some different behavior rather than that of the overall network, while affecting the function of the network too. Up until now, many researchers in various fields have work d on this subject, which has leaded to enormous range of community detection methods with deferent basics and perspectives . However, the community detection problem has not been satisfactorily and comprehensively solved yet , and it is regarded as an open question in complex networks. Of popular approaches to this problem are optimization methods which are based on maximizing modularity or another suitable quality function, dynamical algorithms which employ mobile processes in the network, and spectral methods which use a network matrix in order to find community structures. In this text, finding communities based on the clumpiness matrix in complex networks is investigated and analyzed. In this method, eigenvectors of clumpiness matrix are used to construct a projection space. In such a space, the accumulation of points in branches, which correspond to communities, is observed. These branches are divided by defining a borderline and/or using hierarchical clustering methods, yielding the communities of the network. A physical justification based on the interactional relation between nodes and considering the clumpiness matrix as the Hamiltonian of the system, is presented to explain the performance of the method. Accordingly, the effect of heterogeneity in the community size distribution is discussed. Then, the computational results of the method are presented and its performance on benchmark and real networks is compared with other algorithms. Keywords : Real networks, community structure, random graphs, quality function, spectral analysis, dynamical processes.