Surface air temperatures modelled by ERA-40, ERA-Interim and (NCEP)/(NCAR) reanalysis (NNRP-1) have been compared with observations at 11 synoptic stations in Ireland over the period 1989–2001. The three reanalysis datasets show good agreement with the observed data and with each other. Slopes of the least-squares line to scatter plots of reanalysis data versus observational data show small differences between the three reanalyses, with ERA-40, ERA-Interim and NNRP-1 slopes ranging between (0.79–1.06) ± 0.01, (0.83–1.01) ± 0.01 and (0.76–0.98) ± 0.01, respectively. Summary statistics and the monthly mean temperatures over the 1989–2001 period showed that the reanalyses were significantly warmer in winter than the observations, which resulted in best fit lines with slopes consistently less than unity. ERA-Interim was slightly better than both ERA-40 and NNRP-1 at modelling winter temperatures and it had higher correlation coefficients with the observations. All three reanalyses use different grid sizes and types. Subsequent regridding of the ERA-Interim and NNRP-1 data to the ERA-40 grid showed that the grid difference had no significant influence on the results. Comparison of ERA-Interim and NNRP-1 data with the air temperatures at four marine buoys around the Irish coast for the period 2001–2005 showed that the reanalyses modelled colder winter temperatures than the observations; resulting in best fit lines with slopes consistently greater than unity. The slopes for NNRP-1 and ERA-Interim at the marine buoys, respectively, averaged 1.09 ± 0.04 and 1.10 ± 0.05 while the slopes at the four land stations over the same period averaged 0.87 ± 0.02 and 0.89 ± 0.02, respectively. We believe that this pattern results from the difference in the treatment of land and sea surfaces in the reanalysis datasets.