The proton-resonance frequency (PRF) shift method of MR thermometry provides an easy and practical means of quantitatively monitoring in vivo temperatures for MR image-guided thermal-coagulation therapy. However, reported discrepancies in the numerical value of the PRF-thermal coefficient persist, when measured in a variety of experimental conditions and in different tissue types, both ex vivo and in vivo. In this report, a potential source of variation in the PRF-shift method of thermometry is identified that manifests as a constant incremental phase shift per unit change in temperature that is independent of the echo-time setting, when constructing temperature-sensitive phase images from a gradient-echo pulse sequence. It is proposed that this confounding phase-shift offset arises from thermally induced changes in the electrical conductivity of the material. To this end, it is demonstrated that the MR-derived temperature changes could be in error by as much as 28%, as measured from a simple calibration experiment on freshly excised cow liver. A simple method of overcoming this phase-shift offset is described.