The dynamic nature of web and with the increase of the number of web pages, it is very difficult to search required web pages easily and quickly out of thousands of web pages retrieved by a search engine. The solution to this problem is to classify the web pages according to their genre. Automatic genre identification of web pages has become an important area in web page classification, because it can be used to improve the quality of web search results and also to reduce the search time. In this paper, a Combined Stemming Approach (CSA) is proposed to extract genre relevant words and to classify web pages by genre (non-topical) based on word level and linguistic features. Experiments were performed on 7-genre corpus.