AbstractSurface wave testing is a powerful tool for noninvasive seismic site characterization. It includes a wide variety of active-source and passive-wavefield methods [e.g., spectral analysis of surface waves (SASW), multichannel analysis of surface waves (MASW), microtremor array measurements (MAM)] that can be applied to many different field conditions. The dispersion data collected with all these methods can be inverted to produce (one-dimensional) 1D shear wave velocity (VS) profiles of the subsurface. A critical part of this inversion procedure is the need to develop several trial model parameterizations with different numbers of layers as a means for investigating epistemic uncertainty and nonuniqueness when attempting to fit the experimental dispersion data. This is especially important when a priori information either is not available or does not extend to great enough depths to constrain the inversions. These trial parameterizations are used to vary the number of potential layers present in the subsurface models and to set the acceptable search ranges for depths to boundaries and stiffness in each layer. The inversion results are highly affected by the parameterizations. When performing high-quality inversions, it is important to use different parameterization options to fully explore potential conditions at the site and characterize the epistemic uncertainty of the inversion process. While this practice allows for a robust analysis of the experimental dispersion data and its uncertainty, it also generates many potential subsurface models for the site, which can represent a challenge for engineers when deciding which VS profiles best represent the true subsurface layering for use in subsequent analysis and design (e.g., seismic site response). This paper presents an approach, called the DeltaVs method, for systematically evaluating the depths of layer boundaries across all acceptable models determined from several different inversion layering parameterizations as a means to identify distinct clusters of boundaries. Using these clusters, the number and location of meaningful layer boundaries can be determined for many sites, thereby allowing for the elimination of some parameterization options containing more, or fewer, boundaries and a reduction in epistemic uncertainty. The distributions of layer boundaries within these clusters can then be used to determine statistics for the depths of layer boundaries that are consistent with the epistemic uncertainty of the models produced by the inversion procedure. The DeltaVs method is demonstrated on 12 synthetic ground models as well as one field data set.