In this research, we compare 2-m air temperature from the ERA-Interim reanalysis of the European Centre for Medium-Range Weather Forecasting with 2-m air temperature weather station observations in Cuba, with the goal of evaluating the behaviour and uncertainties of the ERA-Interim data set with respect to station-based observations. Three interpolation methods are used to determine 2-m temperatures from the ERA-Interim data set at the station locations. The differences were analysed utilizing root mean square error (RMSE), mean absolute error (MAE) and bias. The comparison was conducted for daily, monthly and annual time scales, and for the rainy (May–October) and less rainy (November–April) seasons. The best interpolation method is the mean of four grid points method. We find a warm bias in the ERA-Interim reanalysis for most Cuban stations. The smallest differences are at 1800 UTC and the largest differences are at 1200 UTC. All differences are greater than 0.3 K, although many of the stations show differences in the range of 1.5–2.0 K. In some stations the differences are greater than 5.0 K. At the daily scale more than 50% of the stations show significant differences at the 95% confidence level. The differences are caused by the altitude difference between the stations and the nearest grid point of ERA-Interim, the land-sea mask of ERA-Interim and the station location respect to this mask, and by local processes, such as a local breeze. At the monthly scale there are fewer stations with significant differences than for the other time scales. The ERA-Interim reanalysis better represents the surface 2-m temperature for coastal stations than for inland stations. Years with moderate and strong El Niño or La Niña show significant differences between ERA-Interim and observations. The amplitude between the maximum bias and the minimum bias is greater in those years.