In this paper, we propose an relative entropy based method to extract meaningful homonyms. Among different types of ambiguity, homonyms are the lexical ambiguity, which should be disambiguated. Since the corpus study reveals that one sense of homonyms favor predominance in usage, such kind of homonyms needs not to be disambiguated. Therefore, there is need to extract meaningful homonyms for disambiguation. Previous study suggests a method that considers the frequency and the frequency ratios. However, there is no correlation between the frequency and the frequency ratio. What we propose in this paper is to use the relative entropy to extract the proper meaningful homonym lists.