Real-time thermal capacity estimation is not only part of prognostics and health management (PHM) but also determines the dynamic loading capability of the traction transformer. The hot-spot temperature is the vital parameter for real-time thermal capacity estimation. In high-speed railway, the unbalanced loads of the two phases of the traction transformer cause unbalanced winding losses and therefore affect the top-oil temperature and further influence the hot-spot temperature in each phase. In order to explore the dynamic thermal performance of the transformer under unbalanced load (current), a thermal model for calculating the hot-spot temperature was proposed incorporating thermal interaction between two power supply phases. This model treats the winding loss of each phase independently, considering the temperature dependence of winding ohmic losses and eddy current losses. A thermal circuit topology was presented with independent phase windings sharing the same top-oil temperature, where an improved heat transfer correlation was used to account for the nonlinear thermal convection. The proposed model was verified by the factory temperature rise tests and field operation data for the same batch of transformers.