This paper describes an experimental comparison between two standard supervised learning methods, namely Naive Bayes and Exemplar--based classification, on the Word Sense Disambiguation (WSD) problem.