The proton-resonance frequency (PRF) shift method of thermometry has become a promising tool for magnetic resonance image-guided thermal therapies. Although the PRF thermal coefficient has recently been shown to be independent of tissue type when measured ex vivo, significant discrepancy remains on its value for tissues measured in vivo under a variety of experimental conditions. The authors identify a potential source of variation in the PRF thermal coefficient that arises from temperature-induced changes in the volume magnetic susceptibility of tissue and is dependent on the orientation and geometry of the heat-delivery device and its associated heat pattern. This study demonstrates that spatial variations in the apparent PRF thermal coefficient could lead to errors of up to +/-30% in the magnetic resonance estimated temperature change if this effect is ignored.