<p><strong>Abstract.</strong> The global response of air temperature at 2 metre above the surface to the eruptions of Mount Agung in March 1963, El Chich&#243;n in April 1982, and Mount Pinatubo in June 1991 is investigated using 11 global atmospheric reanalysis data sets (JRA-55, JRA-25, MERRA-2, MERRA, ERA-Interim, ERA-40, CFSR, NCEP-NCAR R-1, 20CR version 2c, ERA-20C, and CERA-20C). Multiple linear regression (MLR) is applied to the monthly mean time series of temperature for two periods, 1980&#8211;2010 (for 10 reanalyses) and 1958&#8211;2001 (for six reanalyses), by considering explanatory factors of seasonal harmonics, linear trends, Quasi-Biennial Oscillation (QBO), solar cycle, tropical sea surface temperature (SST) variations in the Pacific, Indian, and Atlantic Oceans, and Arctic SST variations. Empirical orthogonal function (EOF) analysis is applied to these climatic indices to obtain a set of orthogonal indices to be used for the MLR. The residuals of the MLR are used to define the volcanic signals for the three eruptions separately. First, latitudinally averaged time series of the residuals are investigated and compared with the results from previous studies. Then, the geographical distribution of the response during the peak cooling period after each eruption is investigated. In general, different reanalyses show similar geographical patterns of the response, but with the largest differences in the polar regions. The Pinatubo response shows largest average cooling in the 60&#176;&#8201;N&#8211;60&#176;&#8201;S region among the three eruptions, with a peak cooling of 0.10&#8211;0.15&#8201;K. The El Chich&#243;n response shows slightly larger cooling in the NH than in the Southern Hemisphere (SH), while the Agung response shows larger cooling in the SH. These hemispheric differences are consistent with the distribution of stratospheric aerosol optical depth after these eruptions; however, the peak cooling after these two eruptions is comparable in magnitude to unexplained cooling events in other periods without volcanic influence. Other methods in which the MLR model is used with different sets of indices are also tested, and it is found that careful treatment of tropical SST variability is necessary to evaluate the surface response to volcanic eruptions in observations and reanalyses.</p>