Abstract-Common approaches for partitioning cell site planning problems into smaller instances typically suffer from the lack of knowledge about the number of clusters k that is appropriate for the particular problem. When applying an iterative processing over methods such as the k-means algorithm or the related k-medoids approach to determine the optimum k, the computational effort can be higher than solving the original problem instance directly. This is particularly the case if the optimum k is small, which happens most likely in urban environments where the user density and the number of cells are high. In this paper, we propose a graph-based method using minimum cut operations to partition a planning instance automatically into an appropriate number of k sub-problems. We demonstrate the benefits of this approach by numerical evaluation of exemplary application to the problem of planning LTE cell site locations in an urban environment.