Complex networks partitioning consists in identifying denser groups of nodes. This popular research topic has applications in many fields such as biology, social sciences and physics. This led to many different partition algorithms, most of them based on Newman's modularity measure, which estimates the quality of a partition. Until now, these algorithms were tested only on a few real networks or unrealistic artificial ones. In this work, we use the more realistic generative model developed by Lancichinetti et al. to compare seven algorithms: Edge-betweenness, Eigenvector, Fast Greedy, Label Propagation, Markov Clustering, Spin glass and Walktrap. We used normalized mutual information (NMI) to assess their performances. Our results show Spinglass and Walktrap are above the others in terms of quality, while Markov Clustering and Edge-Betweenness also achieve good performance. Additionally, we compared NMI and modularity and observed they are not necessarily related: some algorithms produce better partitions while getting lower modularity.