Satellite remote sensing has revolutionized modern oceanography, providing frequent synoptic-scale information that can be used to deduce ocean processes. However, it is often difficult to extract interpretable patterns from satellite images, as data sets are large and often non-linear. In this methodological paper, we describe the self-organizing map (SOM), a type of artificial neural network adept at pattern identification. The ability of the SOM to extract patterns from a variety of satellite data, including scatterometer and thermal imagery, is illustrated by example. We characterize inter-annual, seasonal and event-scale variability by using the SOM and relate the output to auxillary variables by using a number of techniques that enhance interpretation. Practical recommendations for the fruitful application of SOMs are given. Although the SOM has only rarely been used in oceanography previously, it is a promising applied mathematical tool for pattern extraction from many types of data, especially large and complex satellite data sets.