One might expect that the sea surface temperature in areas close to significant sea ice concentrations is just above freezing (-1.8 deg C for sea water), but it is not necessarily so. Sea ice can be transported into 'warmer' waters, and the upper ocean may be significantly warmed by solar heating before nearby ice melts. Here, we are motivated by the need to quantify the relationships between SST and sea-ice in different regions and seasons for the purposes of climatology. The major steps in obtaining SSTs from thermal satellite imagery are classification (identifying clear-sky-clear-ocean pixels) and retrieval (estimating SST from the observed brightness temperatures). Both are particularly challenging in sea ice zones. Cloud cover is persistent and extensive at 1-km resolution, low-lying fog / cloud in particular has a weak thermal signature, and sun-sea-satellite geometry is often unfavourable for use of reflectance channels. Retrievals of SST need to be unbiased by the extreme atmospheric inversions that occur only in these polar latitudes. We present our work on these challenges using ATSR-2. We take advantage of that sensor's near infrared reflectance channels for improved day-time clear-sea detection, and assess the degree to which the dual-view capability of ATSR-2 is useful in defining robust SST retrievals.