In this paper we present an automatic approach for coastline detection from images which is based on parametric active contours (snakes). Snakes require the definition of an energy functional that reflects the underlying coastline model. As for Antarctica, our application domain, the coastline appearance in the used optical images is heterogeneous. Therefore, a single model does not work equally well in all situations. On the basis of an up-to-date Landsat mosaic three different models are formulated that match a large part of the Antarctic coastline, i.e. the transition from ice shelf to water, from ice shelf to sea ice and from rocky terrain to water. For each of the three different cases the energy terms are optimized based on the radiometric properties of the adjacent regions as well as the curvature and the potential change-rate of the coastline itself. A supervised classification for the three classes ice, water and rocky terrain controls the whole process by choosing the most applicable model for a certain image region. With a view to the practical application the developed approach was integrated into a semiautomatic system, where the human operator supervises the optimization process of the contour and interactively corrects the results if the system fails.