Web page classification differs from traditional text classification due to additional information by Hyper Text Markup Language (HTML) structure and the presence of hyperlinks. While effort was taken to exploit hyperlinks for classification, web pages structured nature is rarely considered. A noticeable HTML documents feature is HTML tags and respective attributes that ensure that HTML documents are viewed in browsers and other user agents. This paper proposes a semantic-based feature selection to improve web pages search and retrieval over large document repositories. Web page classification using HTML tags is evaluated using the 4 Universities Dataset. The features are classified using Proposed Neural Network. The experimental results show improved precision and recall with the presented method.