The invention discloses an incremental learning multi-level binary-classification method of scientific news. The method includes the steps that article titles, article contents and key words in the property of news are utilized in combination with a text weighted method and a text similarity calculation method under a vector space model, preprocessing and feature weighing are firstly conducted on marked information and full-text information collected in a marked news documents; an intermediate result is calculated and stored; then, the similarity between a new text and scientific news classification and non-scientific news classification is calculated through cosine similarity on the aspects of feature information and a full text, and accordingly the classification of the new text is judged. Sensitiveness to scientific and technical vocabularies through the multi-level judgment method and the incremental learning method, the number of texts of news unrelated to the scientific news can be reduced through the binary-classification method, and thus the text multi-classification accuracy is improved. The incremental learning multi-level binary-classification method of the scientific news is used for improving the use value of news information extracted from Web pages and improving the classification accuracy of the scientific news.