We have applied multiscale deformable-layer tomography (DLT) to build a laterally varying velocity model, using a single-offset vertical seismic profile (VSP) data set acquired for a salt proximity survey in southern Texas. The purpose of the VSP survey is to delineate the 2D salt flank using the P-wave reflections. Previous study has identified an anhydrate layer as the cap rock of the salt dome. The large impedance contrasts of this anhydrite layer generate strong downgoing P (sediment)-S (anhydrite)-P (salt) waves recorded by downhole geophones. Incidentally, the P-S-P-waves have traveltimes similar to those of the P-wave salt flank reflections, thus contaminating the imaging of the salt flank. Identifying shear-mode contamination requires an accurate velocity model of anhydrite. However, the extremely poor coverage of the single-offset VSP greatly challenges tomographic techniques to determine the lateral velocity variation. We tackle this problem using multiscale DLT, which characterizes the velocity field by a set of deformable layers. We constrain the layer velocities using the check-shot data and invert for the geometric variation. The inverted model indicates that the anhydrite layer has a “thick-thin-thick” lateral variation with offset, and the S-wave in the anhydrite layer helps in imaging the P-S-P-waves along the well track. The estimated anhydrite layer geometry is validated by the kinematic accuracies of P-waves in the data domain and P-S-P-waves in the image domain. Some in-salt dipping structures are determined by multiscale DLT as well. This field data example indicates that multiscale DLT is feasible for estimating velocities using VSP data of the single-offset situation. An accurate velocity model is the key for modeling and adaptive subtraction of the shear-mode contamination related to the salt geometry.