Ground thawing and freezing depths (GTFDs) strongly influence the hydrology and energy balances of permafrost regions. Current methods to simulate GTFD differ in algorithm type, soil parameterization, representation of latent heat, and unfrozen water content. In this study, five algorithms (one semiempirical, two analytical, and two numerical), three soil thermal conductivity parameterizations, and three unfrozen water parameterizations were evaluated against detailed field measurements at four field sites in Canada's discontinuous permafrost region. Key findings include: (1) de Vries' parameterization is recommended to determine the thermal conductivity in permafrost soils; (2) the three unfrozen water parameterization methods exhibited little difference in terms of GTFD simulations, yet the segmented linear function is the simplest to be implemented; (3) the semiempirical algorithm reasonably simulates thawing at permafrost sites and freezing at seasonal frost sites with site‐specific calibration. However, large interannual and intersite variations in calibration coefficients limit its applicability for dynamic analysis; (4) when driven by surface forcing, analytical algorithms performed marginally better than the semiempirical algorithm. The inclusion of bottom forcing improved analytical algorithm performance, yet their results were still poor compared with those achieved by numerical algorithms; (5) when supplied with the optimal inputs, soil parameterizations, and model configurations, the numerical algorithm with latent heat treated as an apparent heat capacity achieved the best GTFD simulations among all algorithms at all sites. Replacing the observed bottom temperature with a zero heat flux boundary condition did not significantly reduce simulation accuracy, while assuming a saturated profile caused large errors at several sites.