Subsurface mean annual ground temperatures in frozen-ground regions typically differ from those at the ground surface because the thermal conductivity of the seasonal freeze–thaw layer changes between its thawed and frozen states. Mean annual ground temperature at the base of the freeze–thaw layer is therefore a more representative measure of the thermal state of frozen ground than mean annual ground surface temperature. A widely used analytical model for estimating the mean annual ground temperature at the base of the freeze–thaw layer from ground surface thawing and freezing indices, hereafter referred to as the thermal offset model, was introduced over three decades ago. However, its physical basis, assumptions, boundary conditions, and limits of applicability have remained poorly understood because previous publications have either presented the model without a derivation, derived it empirically, or provided mathematically complicated and cumbersome analytical derivations. This paper provides a new and simple analytical derivation of the model using the Stefan equation for seasonal thawing and freezing. This derivation is more physically intuitive and easier to understand and explains why the model takes different forms under permafrost and seasonally frozen ground conditions, clarifies how it is controlled by the ratio of effective thawed to frozen ground thermal conductivities, and reveals how it relates to the thickness of the freeze–thaw layer and latent heat. It also shows that, under sinusoidal ground surface temperature variations, the model can be reformulated in terms of the mean annual ground surface temperature and annual ground surface temperature amplitude. Lastly, it addresses several terminological ambiguities that have persisted since the model was introduced. Overall, the paper shows that the thermal offset model is physically based but strongly simplified, clarifies its behaviour and limits of applicability, and provides a basis for its further development.