Earthquake-based models of slab geometry are limited by the distribution of earthquakes within a subducting slab, which is often heterogeneous. The fast seismic velocity signature of slabs in tomography studies is independent of the distribution of earthquakes within the slab, providing a critical constraint on slab geometry when earthquakes are absent. In order to utilize this constraint, researchers typically hand-contour images of subducting slabs in tomography models, leading to a subjective final slab model. With this paper, we present an automated procedure for extracting slab geometry from teleseismic tomography volumes that limits this subjectivity and provides constraints on the structure of aseismic segments of slabs. This procedure is designed as a complement to earthquake-based slab models rather than as a replacement, which can help to broaden the extent of existing subduction zone geometry databases.