North Atlantic sea surface temperature data from the Comprehensive Ocean-Atmosphere Data Set were used to investigate the behavior of temperature anomalies on multiple-year timescales during the period 1948--1992. Monthly anomaly time series for each 2¿ square from the equator to 70 ¿N were low-pass filtered at 4 years and normalized by the local standard deviation. Attention is focused on the extreme events, the upper and lower deciles, of the anomaly time series. A 45-year sequence of January maps shows the already familiar phenomena of generally cold conditions prior to 1951, a long warm interval from 1951 through 1967, and again a cold period from 1968 through 1977. The years 1978 through 1982 were largely devoid of persistent strong anomalies, but moderate cold conditions returned during 1983--1986. Warm conditions dominated the North Atlantic from 1987 onward. Within these thermal epochs, however, a total of five cold anomaly features and nine warm anomaly features have been identified. These features have individual lifetimes of 3 to 10 years. A typical size is 20¿ of latitude or longitude, but they range from barely detectable to spanning the width of the basin, the latter especially in lower latitudes. Most of the anomalies move long distances along certain preferred paths. These paths generally follow the routes of the subarctic and subtropical gyres. Anomalies originating off North America along the boundary between the gyres move northeastward toward the Norwegian Sea along the approximate route of the North Atlantic Current. Midlatitude anomalies originating at the eastern boundary tend to spread both northward and southward along the coast. The speed of these movements (1--3 km d-1) is generally less than the expected speed of the near-surface ocean circulation. Simple ideas about the effects of beta dynamics and air-sea heat exchanges are briefly considered but do not provide a satisfactory explanation for the movements of the anomalies. The long timescale of these extreme events and the continuity of their movements suggest a useful degree of predictability of sea surface temperature based on persistence and propagation of features. ¿ American Geophysical Union 1996