We analyze surface air temperature datasets simulated by a coupled climate model forced with different external forcings, to diagnose the relative importance of these forcings to the observed warming in the early 20th century. The geographical distribution of linear temperature trends in the simulations forced only by natural contributions (volcanic eruptions and solar variability) shows better agreement with observed trends than that does the simulations forced only by well-mixed greenhouse gases. Using an optimal fingerprinting technique we robustly detect a significant natural contribution to the early 20th century warming. In addition, the amplitude of our simulated natural signal is consistent with the observations. Over the same period, however, we could not detect a greenhouse gas signal in the observed surface temperature in the presence of the external natural forcings. Hence our analysis suggests that external natural factors caused more warming in the early 20th century than anthropogenic factors.