We identify the major spatial patterns of variability in the global surface air temperature data set using empirical orthogonal analysis. We rotate the major components to simplify physical interpretation. Of the five patterns which account for the largest proportion of the total variance in the data set, three are global in extent and two are continental in scale. The pattern accounting for most of the variance in the data set indicates a global change in temperature, affecting most regions in the same sense but most marked in lower latitudes. The second most important pattern is a measure of the El Ni¿o-Southern Oscillation phenomenon. The third pattern represents a contrast in temperature between the northern and southern oceans and is shown to be related to Sahel rainfall variations, confirming, though with reservations, the results of previous work. The fourth pattern is an index of the temperature effects of the North Atlantic Oscillation. The fifth pattern is a measure of temperature variations over Siberia. It is shown that this component is related to an atmospheric circulation fluctuation affecting, among other things, the strength of the depression track over the Barents and Kara Seas and neighboring areas. We term this process the Euro-Siberian Oscillation. The effects of changing data coverage are explored, and it is shown that patterns with a strong maritime component are particularly susceptible to this form of inhomogeneity. Finally, potential causal mechanisms, such as volcanic pollution of the atmosphere, are investigated, and the evolution over time of the spatial response to these forcing factors is defined using the empirical orthogonal functions. The analysis demonstrates that empirical orthogonal functions representing the spatial patterns of temperature change provide a concise means of monitoring climate trends and can assist identification of the causes of recent climate change as well as supporting climate model validation and assessment of the representativeness and reliability of climate reconstructions based on proxy indicators such as tree ring data. ¿ 1999 American Geophysical Union