The internal variability in a 1000-yr control simulation with the coupled atmosphere–ocean global climate model ECHO-G is analysed using near-surface temperature, precipitation and mean sea level pressure variables, and is compared with observations and other coupled climate model simulations. ECHO-G requires annual mean flux adjustments for heat and freshwater in order to simulate no significant climate drift for 1000 yr, but no flux adjustments for momentum. The ECHO-G control run captures well most aspects of the observed seasonal and annual climatology and of the interannual to decadal variability of the three variables. Model biases are very close to those in ECHAM4 (atmospheric component of ECHO-G) stand-alone integrations with prescribed observed sea surface temperature. A trend comparison between observed and modelled near-surface temperatures shows that the observed near-surface global warming is larger than internal variability produced by ECHO-G, supporting previous studies. The simulated global mean near-surface temperatures, however, show a 2-yr spectral peak which is linked with a strong biennial bias of energy in the El Niño Southern Oscillation signal. Consequently, the interannual variability (3–9 yr) is underestimated.