We examine the available observational record of Arctic sea-ice extent for a possible anthropogenic signal. In the light of the discrepancy between climate model simulations and observed sea-ice retreat, it is desirable not having to rely on climate model simulations in the analysis of a possible anthropogenic signal in the recent retreat. We therefore focus in our analysis on the available observational record and derive an estimate of natural variability of Arctic sea ice that compares very well with pre-industrial model simulations. Using this natural variability to analyze the recent retreat, we find that even with very conservative estimators the sea-ice minimum that was reached in 2007 would occur far less than once in 10 000 years without an anthropogenic cause. We also discuss possible inhomogeneities in the available observational record that might cause an overestimate of the recent retreat.