We compare the performance of five land surface models (Chameleon Surface Model (CHASM), Noah, Community Land Model (CLM), Variable Infiltration Capacity Model (VIC), European Centre for Medium Range Weather Forecasts (ECMWF)), in the simulation of hydrological processes across the terrestrial Arctic drainage system for the period 1980–2001. The models represent a wide range of model physics, particularly with respect to high-latitude processes, and are forced with surface meteorology derived from the ERA-40 reanalysis. Models offer great potential for enlightenment regarding large-scale hydrology in this poorly observed region; thus our objective is to assess the ability of the models to capture various aspects of pan-Arctic hydrology as well as identify those features that contain the largest uncertainty. Results reveal up to a 30% difference in annual partitioning of precipitation between evaporation and runoff with major Arctic watersheds such as the Lena. Capturing the correct base flow of the large rivers is a consistent problem. The model hydrographs are often out of phase, peaking too early in comparison to observations. However, allowing for a large uptake in soil moisture as well as moisture movement during frozen periods alleviates this discrepancy. A negative correlation exists between models and observations for annual runoff time series over the Yenesi basin, apparently mostly because of inconsistencies in the input data. Compared to station data, all models produce similar errors in snow water equivalent; yet they differ widely in their snow regimes in terms of snowfall quantity, estimated snow depths, and most importantly, sublimation rates. Additionally, model albedo is consistently higher than observations in the presence of snow. No single model is the best or worst performing when compared to a range of observations.