Text summary is a small article to give a user an overview or a key-information about the content of a document or document collection. Especially, summaries are useful to retrieve some pieces of information or to cluster a set of documents when the size of set is large. Summaries are categorized into abstract, a human-made summary and extract, a part of original document. The purpose of human-made abstracts, generally produced by the author herself, is to give reader a short and coherent impression of the main idea of the article, but this purpose is hard to be achieved by a machine system. However, relevant and effective summaries can be generated by extracted verbatim from original articles. For many purposes, coherence is not a crucial point, but relevance certainly is. Relevance of a sentence in a document is determined with statistical information or text understanding knowledge. We investigate a simple hyper-information feature, named theme link, which has both of statistical information and text understanding knowledge. A theme link consists of a phenomenon term which describes a class of events and fact term which specifies a class of events and makes it a event. This theme link is a good feature to express a event of news article, A summary, which is composed of highest weighted sentences by theme link, expresses the event more than by other feature such as tf-idf and Itc. Moreover, we can see that a summarization system with theme link can reduce the redundancy involved in result effectively.