The mapping of the snow cover from optical remote sensing is a largely investigated topic that still present some challenges especially in mountainous areas when high resolution time series are exploited. Providing reliable maps in all the acquisition conditions is limited by topographic effects such as shadows, sunglint on snow and atmospheric disturbances. If these effects are not corrected, the state of the art methods are generally failing. Support vector machine (SVM) proved to be an effective tool for dealing with classification problems, even when data are noisy and the classes are not linearly separable. In this work, we propose an unsupervised approach to snow cover fraction retrieval that collect “snow” and “snow free” samples to train a model tailored for the specific acquisition conditions of each image. The method has been applied in different test sites in the Alps. For a quantitative evaluation of the result, we used a snow maps derived by very high resolution images acquired in challenging situations of fractional snow cover and low sun elevation.