The high latitudes form an important component of the Earth’s carbon cycle. It is therefore important to capture this in Earth System Models (ESM’s). However, most carbon-cycle development and evaluation in ESM’s focuses on lower latitudes, and therefore there is an urgent need to address Arctic carbon-cycle processes. Here, we run land-surface schemes from ESM’s at the site level at various Arctic sites, performing a detailed evaluation of the carbon dynamics in the models. They are process-based models, and therefore point-scale evaluations contribute directly towards improving the large-scale results. The sites chosen for the simulations are the five principal field sites from the recently-concluded EU project PAGE21. This gives the distinct advantage that detailed data are available. In particular, data on the physical state of the climate and permafrost at these sites, and large datasets of soil carbon stocks and fluxes. The sites cover a range from low Arctic discontinuous permafrost to high Arctic desert, and a range of soil types from thick peat to mineral soils with little organic matter. The models involved are land surface schemes from three European ESM’s: UKESM (UK), IPSL (France) and MPI-ESM (Germany). The models all have improved process representation as part of PAGE21. The simulations are first compared with physical observations from the sites: specifically snow depth, soil temperature, soil moisture and maximum thaw depth. All models capture the physics with a reasonable accuracy, and certainly capture the major differences between sites, with a few exceptions. In particular, we see the importance of simulating the physical properties of the soil organic layer. Comparing simulated soil organic carbon with observations shows the importance of including vertical soil carbon profiles. In one model this is not represented, which results in a failure to capture the differences in soil carbon in different physical conditions. Including cryoturbation mixing is key to simulating the vertical soil organic carbon profile. When vertical mixing is included, the profile of soil organic carbon at mineral soil sites matches very well with observations. However, none of the models are able to simulate the correct profile at sites with organic soils, highlighting the need for further process representation of peat accumulation. Finally, the land-atmosphere carbon fluxes are assessed using different observations, and we discuss the meaning of these measurements in terms of the land surface model output and how they can be most usefully compared. The simulation of carbon fluxes depends on every aspect of the models: The physical state, the soil carbon stocks and most importantly, the vegetation. Large errors result from the models growing the wrong type of vegetation or no vegetation at all, as tundra vegetation types are not represented in two of the three models. Focussed work is required to better represent Arctic vegetation in such models, and our results highlight the next steps to take. The observational datasets are more detailed than those used in past studies and this work will be used both to facilitate and to justify the development of Arctic carbon cycle processes in Earth system models.