The rate of global mean surface temperature (GMST) warming has slowed this century despite the increasing concentrations of greenhouse gases. Climate model experiments [1][2][3][4] show that this slowdown was largely driven by a negative phase of the Pacific Decadal Oscillation (PDO), with a smaller external contribution from solar variability, and volcanic and anthropogenic aerosols 5,6 . The prevailing view is that this negative PDO occurred through internal variability 7-11 . However, here we show that coupled models from the Fifth Coupled Model Intercomparison Project robustly simulate a negative PDO in response to anthropogenic aerosols implying a potentially important role for external human influences. The recovery from the eruption of Mount Pinatubo in 1991 also contributed to the slowdown in GMST trends. Our results suggest that a slowdown in GMST trends could have been predicted in advance, and that future reduction of anthropogenic aerosol emissions, particularly from China, would promote a positive PDO and increased GMST trends over the coming years. Furthermore, the overestimation of the magnitude of recent warming by models is substantially reduced by using detection and attribution analysis to rescale their response to external factors, especially cooling following volcanic eruptions. Improved understanding of external influences on climate is therefore crucial to constrain near-term climate predictions.The recent slowdown in surface temperature warming is clearly seen in observed time series of 15-year global mean surface temperature (GMST) trends 8,9 that reached a peak in the 15-year period 1992-2006 followed by a sharp decline (black curves in Fig. 1a). This is seen in all of the leading observational data sets despite recent claims that the slowdown is eliminated by corrections to sea surface temperature biases 12 . A similar peak and decline is also simulated by all coupled models from the Fifth Coupled Model Intercomparison Project (CMIP5, coloured curves in Fig. 1a, with thick red curve showing the ensemble mean, see Methods and Supplementary Table 1). Similar results are obtained for different trend lengths ( Supplementary Fig. 1). The modelled trends are generally larger than observed. This is an important point that will be addressed later, but we first investigate the cause of the simulated peak and decline in GMST trends.Since internal variability is unlikely to be in phase in the CMIP5 model simulations, any common signals are likely to be externally forced. We therefore investigate additional CMIP5 model simulations that are forced by different combinations of external factors (Supplementary Table 1 and Methods). These show that the sharp peak in the trend in the period 1992-2006 is caused by natural factors (Nat, green curve in Fig. 1b), and in particular the eruption of Mount Pinatubo as pointed out previously 9,13 . The reason is straightforward: the eruption of Mount Pinatubo in 1991 caused −0.6 −0.4 −0.2 −0.0 0.2 0.4 0.6 0.8 a −0.2 0.0 0.2 0.4 b Obs. All GHGs Nat. Aero. End year...