Valid estimates of sea surface temperature (SST) from the advanced very high resolution radiometer (AVHRR) are critically dependent upon the identification and removal of clouds from parent images. Previous cloud-screening algorithms have demonstrated varying degrees of success, the effectiveness of the particular method used being critically dependent upon cloud type. Contamination from cirrus clouds, for example, has been a major deficiency in previous methods. A robust new algorithm for cloud masking has now been developed which is very effective at removing cloud contamination from images regardless of cloud type. The method, which evaluates every pixel in the image, is statistically reproducible, computationally efficient, and requires no subjective inputs or knowledge of cloud type. The results also show that the algorithm is not regionally specific. The basis for this new method is presented and applications to global AVHRR data, including comparisons with previously used methods, are shown. ¿ American Geophysical Union 1990