The task of entity linking consists in disambiguating named entities occurring in textual data by linking them to an identifier in a knowledge base that represents the real-world entity they denote. We present Weasel, a novel approach that is based on a combination of different features that is trained using a Support Vector Machine. We compare our approach to state-of-the-art tools such as FOX and DBpedia spotlight, showing that it outperforms both on the AIDA/CoNLL dataset and provides comparable results on the KORE50 dataset.