In this paper we propose and discuss a method for Word Sense Disambiguation. A Lexicon approach is presented based on the use of the WordNet. More precisely, the context and the senses of the ambiguous word are represented as vectors of weighted terms, in a vector space model, using WordNet definitions and the rich hypernymy relations. Calculating the conditional probabilities (relative frequencies) for these terms we can measure the similarity of the target word with a sense. Hence, the ambiguous word in the context is assigned to the most similar sense. Our algorithm does not need any training and is tested on the entire Semantic Concordance Corpus (Semcor). The estimated performance of the algorithm is 78,13% .