We attempt the removal of the large-scale signal associated with one important component of climate variability, the El Niño-Southern Oscillation (ENSO) phenomenon, from the global surface air temperature data set in order to facilitate analysis of other potential causes of climate change. Previous attempts to remove the ENSO signal from climate records at the global scale have been based on relatively simple regression analysis. We use empirical orthogonal function analysis to identify characteristic spatial patterns of change in the global surface air temperature field and thus define the ENSO signal. The approach successfully identifies two ENSO-related components of the variability in the global data set, and the removal process results in a significant reduction in variance. Evaluation of the process indicates that attention should be paid to seasonality in the ENSO signal. Moreover, it is clear that other aspects of the statistical generalization inherent in the approach have resulted in a residual ENSO signal in the data set. The remaining signal does, however, appear to be regional in nature and/or only marked in the case of particular ENSO events. We conclude that an approach based on empirical orthogonal function analysis presents an effective means of isolating and removing specific climate signals.