We demonstrate how structural parcellation can be implemented using shortest-path tractography, thereby addressing some of the shortcomings of the conventional approaches. In particular, our algorithm quantifies, via p-values, the confidence that a voxel in the parcellated region is connected to each cortical target region. Calculation of these statistical measures is derived from a rank-based test on the histogram of tract-based scores from all the shortest paths found between the source voxel and each voxel within the target region. Using data from the Human Connectome Project, we show that parcellation of the thalamus results in p-value maps that are spatially coherent across subjects. Comparing to the state-of-the-art parcellation of Behrens et al. [1], we observe some agreement, but the soft segmentation exhibits better stability for voxels connected to multiple target regions.