Inorganic contaminants, including potentially toxic metals (PTMs), originating from un-reclaimed abandoned mine areas may accumulate in soils and present significant distress to environmental and public health. The ability to generate realistic spatial distribution models of such contamination is important for risk assessment and remedial planning of sites where this has oc-curred. This study evaluated the prediction accuracy of optimized ordinary kriging compared to spatial regression-informed cokriging for PTMs (Zn, Mn, Cu, Pb, and Cd) in soils near abandoned mines in Bumpus Cove, Tennessee, USA. Cokriging variables and neighborhood sizes were system-atically selected from prior statistical analyses based on the association with PTM transport and soil physico-chemical properties (soil texture, moisture content, bulk density, pH, cation exchange capacity (CEC), and total organic carbon (TOC)). A log transform was applied to fit the frequency histograms to a normal distribution. Superior models were chosen based on six diagnostics (ME, RMS, MES, RMSS, ASE, and ASE-RMS), which produced mixed results. Cokriging models were preferred for Mn, Zn, Cu, and Cd, whereas ordinary kriging yielded better model results for Pb. This study determined that the preliminary process of developing spatial regression models, thus enabling the selection of contributing soil properties, can improve the interpolation accuracy of PTMs in abandoned mine sites.