The ability to correctly identify landform features from remotely sensed imagery is largely determined by pixel size. This spatial scale element is a particularly important consideration for glacial geomorphological feature identification when remote sensing data are used for mapping and palaeoenvironmental reconstruction in mountain environments. It is important to be able to clearly delineate the boundaries of glacial landforms. However, in common with other phenomena, such features often possess indeterminate boundaries and many geomorphometric changes occur over short distances. Thus, the use of conventional hard classification techniques may not always be appropriate in glacial terrain mapping where investigators are concerned with individual feature identification. In such situations the ability to examine sub-pixel scale information by using soft classifiers is potentially more useful. This paper examines the value of sub-pixel data interpretation derived from supervised and unsupervised fuzzy modelling techniques for the mapping and interpretation of glacial terrains for the wider purposes of glacial reconstruction in the Pindus Mountains of Northwest Greece. This work is part of a larger study involving field-based investigation of the glacial sediments and landforms. Emphasis has been given to the effective delineation of features from 20 m resolution SPOT HRV imagery.