The surface frost number model, a statistical-empirical method, is able to simulate and predict frozen soil distribution, which has been applied successfully in high latitude regions. However, owing to lack of physical meaning, it should be carefully treated for applying in other regions. For example, it is unclear whether or not a threshold F=0.5 is sufficient to delineate frozen soil distribution in high altitude regions such as the Tibetan Plateau. In this research, the surface frost number model is re-derived from an explicit physical formulation, the classical Stephan's solution. A new variable E is introduced to explicitly reflect the thermal effects of frozen and thawed states. The original Nelson's surface frost number model can be implemented as a specific instance of E=1. A set of simulations on the Tibetan Plateau with different E were carried out using observed surface soil temperature data. It is found that this model can depict major characteristics of frozen soil distribution over the plateau. A comparative analysis shows that E, representing soil type, has obvious impact on frozen soil distribution, in particular in the northern and southern permafrost boundaries of the plateau. It could be expected that a better simulation can be obtained using a calibrated parameter E.