The present invention provides a method for document classification, especially an adaptive learning method for document classification. The document includes a plurality of feature words. The method includes steps of calculating a plurality of similarities between the document and a categorical basic knowledge; calculating a first ratio of a first largest similarity to a second largest similarity of the plurality of similarities; storing the feature words of the document as an extensive categorical knowledge when the first ratio is larger than a first threshold value; and updating the categorical basic knowledge by using the extensive categorical knowledge.