The automation of pattern recognition in the field of remote sensing involves several preprocessing steps to remove noise and nonuseful data. When infrared data are used to obtain either ocean or land information, cloud pixels must first be identified and eliminated from the image, because cloud contamination is the main producer of errors in deriving sea surface temperatures from remotely sensed data. Cloud masking is usually tackled as a statistical classification problem using threshold or texture-based information from satellite scenes. We attempt to construct an automatic cloud-masking system which uses heuristic knowledge about cloud features in Advanced Very High Resolution Radiometer scenes and artificial neural networks as classifiers. This system could be used as a preprocessing step in a future automatic oceanic feature identification system now being developed for the North Atlantic Ocean. The system has been compared with other traditional cloud mask methods to determine its accuracy.