The depth and duration of soil freezing have important implications for the hydrology, biology, and chemistry of ecosystems. Four existing soil models capable of simulating subfreezing soil temperatures and frost depth were evaluated for their ability to predict the depth and timing of soil frost at sites in North America. The evaluation was carried out by comparing model simulations with field data collected in Alaska, characterized by a Cryaquept with grass cover, and in Minnesota, characterized by a Haploboroll with corn stubble. The SHAW and SOIL models employ a finite difference solution to assess heat flow in the soil profile. Both models predicted frost depth with reasonable accuracy, at least when the simulated snow depth agreed with the recorded snow depth. The Benoit and Gusev models assess frost depth by balancing heat fluxes within the soil profile. These models generally overpredicted frost depth. The chief advantages of the simpler Benoit and Gusev models are the fewer data requirements and faster execution times compared with the SHAW and SOIL models. The latter two models, however, include provisions to reduce the data requirements by utilizing default data values in the simulation. The greater accuracy attained using the more sophisticated modern computer models may warrant their use for site-specific environmental applications. This study illustrates the difficulty of simulating snow cover, and, therefore, soil frost penetration, accurately.