We present a novel, graph-based fibre tractography method, where the graph model can integrate any diffusion representation provided as an ODF. Tractography is performed by computing all possible shortest paths between regions of interest. For each path a confidence value is computed to assess its quality. Voxels are scored by the average value of paths they lie upon. Results for three different known tracts show the robustness of the method over five healthy subjects and a high overlap with tract atlases. The region based approach is especially useful for creating brain connectivity graphs while not suffering from path length dependencies.