Abstract. It is important to correctly simulate permafrost in global climate models, since the stored carbon represents the source of a potentially important climate feedback. This carbon feedback depends on the physical state of the permafrost. We have therefore included improved physical permafrost processes in JULES (Joint UK Land Environment Simulator), which is the land-surface scheme used in the Hadley Centre climate models. The thermal and hydraulic properties of the soil were modified to account for the presence of organic matter, and the insulating effects of a surface layer of moss were added, allowing for fractional moss cover. These processes are particularly relevant in permafrost zones. We also simulate a higher-resolution soil column and deeper soil, and include an additional thermal column at the base of the soil to represent bedrock. In addition, the snow scheme was improved to allow it to run with arbitrarily thin layers. Point-site simulations at Samoylov Island, Siberia, show that the model is now able to simulate soil temperatures and thaw depth much closer to the observations. The root mean square error for the near-surface soil temperatures reduces by approximately 30%, and the active layer thickness is reduced from being over 1 m too deep to within 0.1 m of the observed active layer thickness. All of the model improvements contribute to improving the simulations, with organic matter having the single greatest impact. A new method is used to estimate active layer depth more accurately using the fraction of unfrozen water. Soil hydrology and snow are investigated further by holding the soil moisture fixed and adjusting the parameters to make the soil moisture and snow density match better with observations. The root mean square error in near-surface soil temperatures is reduced by a further 20% as a result.