Following the work of Inderjit S. Dhillon, this paper presents the document clustering problem as a graph partitioning problem. To solve this problem, we use the fusion-fission metaheuristic which has already been applied to several graph partitioning problems. The results obtained with the fusion-fission algorithm are better than those of Graclus, a state of the art graph partitioning software created by Inderjit S. Dhillon. But surprisingly, regarding Inderjit S. Dhillon's objective function, these results are also always better than those of the real partitioning of the documents. This unexpected fact incite us to conclude that Inderjit S. Dhillon's method to convert a document clustering problem into a graph partitioning problem is wanting. Replacing the normalized cut objective function by an another objective function should minimized this problem. However, we do not suggest yet a new objective function more efficient than the normalized cut.