When a large number of networks is observed, we may wish to identify groups of networks with similar topology. This is a challenging task as networks are complex objects and of varying size and thus difficult to compare. We propose a statistical model-based approach to partition a collection of observed networks into a finite number of homogeneous clusters. This is done by a mixture of stochastic block models. Moreover, we propose a greedy agglomerative algorithm based on the integrated classification likelihood to perform the clustering. We present results of our method obtained for a collection of foodwebs in ecology. We illustrate that the method provides relevant clusterings and that a hierarchy of the clusters is automatically provided by the algorithm. In addition, the estimated model parameters are highly interpretable and useful in practice.