An empirical orthogonal function analysis has been performed on monthly mean sea surface temperature observations in the Atlantic Ocean between 70°N and 30°S for the years 1949–69. This method derives the most important ‘modes’ of variation, defined as those functions which explain the largest total variance in the data field. Such modes are calculated for two sets of temperature deviations in which the annual cycle is both included and excluded. For the seasonal case, only four functions explain more than 90% of the total variance, whereas in the nonseasonal case, ten functions explain less than 50% of the total. the most important seasonal mode indicates lower minimum temperatures in winter in the later years of study. the dominant nonseasonal function shows a cooling trend starting in early 1951. Nonseasonal functions of the tropical region alone suggest large variations, having a characteristic period of a few months.