AbstractDecadal climate prediction is a challenging aspect of climate research. It has been and will be tackled by various modeling groups. This study proposes a simple empirical forecasting system for the near-surface temperature that can be used as a benchmark for climate predictions obtained from atmosphere–ocean GCMs (AOGCMs). It is assumed that the temperature time series can be decomposed into components related to external forcing and internal variability. The considered external forcing consists of the atmospheric CO2 concentration. Separation of the two components is achieved by using the Intergovernmental Panel on Climate Change Fourth Assessment Report (IPCC AR4) twentieth-century integrations. Temperature anomalies due to changing external forcing are described by a linear regression onto the forcing. The future evolution of the external forcing that is needed for predictions is approximated by a linear extrapolation of the forcing prior to the initial time. Temperature anomalies owing to the internal variability are described by an autoregressive model. An evaluation of hindcast experiments shows that the empirical model has a cross-validated correlation skill of 0.84 and a cross-validated rms error of 0.12 K in hindcasting global-mean temperature anomalies 10 years ahead.