Abstract The recent satellite magnetic missions, combined with high quality ground observatory measurements, have provided excellent data for main field modelling. Four different groups submitted seven main-field and eight secular-variation candidate models for IGRF-10. These candidate models were evaluated using several different strategies. Comparing models with independent data was found to be difficult. Valuable information was gained by mapping model differences, computing root mean square differences between all pairs of models and between models and the common mean, and by studying power spectra and azimuthal distributions of coefficient power. The resulting adopted IGRF main-field model for 2005.0, an average of three selected candidate models, is estimated to have a formal root mean square error over the Earth’s surface of only 5 nT, though it is likely that the actual error is somewhat larger than this. Due to the inherent uncertainty in secular variation forecasts, the corresponding error of the adopted secular-variation model for 2005.0–2010.0, an average of four selected candidate models, is estimated at 20 nT/a.