WSD (word sense disambiguation) based on supervised machine learning made a great progress, but it is hard to deal with large-scale WSD because of its 慴ig?labor cost. An unsupervised WSD method is provided in this paper to solve this problem. Only under the knowledge database of sense-words, this method formulates the sense-words and polysemous words in vector space, and based on k-NN (k=1) it calculates the similarity between them to disambiguate polysemous words. The average accuracy is 83.13% for 10 polysemous words in open test by this method.