This article advocates the use of lexical knowledge and semantics to improve the accuracy of information retrieval. A system of measuring text similarity is developed, which attempts to integrate the meaning of texts into the similarity measure. The work hinges on the organization of the synsets in the WordNet according to the semantic relations of hypernymy/hyponymy, metronymy/holonymy and antonymy. A unique measure using link distance between the words in a subgraph of the WordNet has been evolved. This needs word sense disambiguation which in itself is a complex problem. We have developed an algorithm for word sense disambiguation exploiting again the structure of the WordNet. The results support our intuition that including semantics in the measurement of similarity has great promise.