Sea ice concentration is a fundamental property of the Arctic ice-ocean-atmosphere system reflecting both dynamics and thermodynamics. Concentration integrates across space and time and is useful for characterizing both observed and numerically simulated systems. Concentration is reasonably well measured by remote sensing, and several high-quality sea ice concentration data sets exist beginning with the satellite era. In this paper we examine the simulated sea ice concentration from nine ice-ocean numerical models that are part of the coordinated experiments of the Arctic Ocean Model Intercomparison Project (AOMIP). Spatial patterns of means and differences between models and observations, and among models, are compared for a multiyear record and for the September sea ice minimum. Interannual variations are assessed on data with monthly climatology removed. As a proxy for the annual cycle of open water for each model, the total areas with concentration less than 10% are compared among models. Mean ice statistics are computed for grid points with greater than 1% and greater than 10% concentrations. The results show that the models have similar characteristics for the winter months when 100% cover is produced, and most models reproduce an observed minimum in sea ice concentration for 1990. The compared observational data sets use the NASA Team algorithm (Goddard Space Flight Center data, the adjusted or Walsh data, and the Hadley Centre data) and the Bootstrap algorithm. Variability in sea ice concentration is less among the four observational records than among models. Copyright 2007 by the American Geophysical Union.