Considering that each sentence element of a sentence or clause plays an important role in describing case and (or) object in documents, a feature extraction method based on sentence element is proposed in this paper. The method can extract feature terms from documents effectively and weight them accurately. It first extracts sentence elements from dependency relationships, and then selects and weights the terms according to the sentence elements. Experimental results on a public dataset prove the feasibility of our approach and demonstrate its advantage to feature extraction method based on part of speech.