Spam emails has brought great distress to people and most of them spread in the manner of embedded spam images, which makes the traditional filtering difficult to detect. With the efforts of many scholars, spam recognition rate is getting better and better, but some researches ignore the importance of reducing the error rate. Thus an improved method of multiple features fusion is proposed in this paper, which combines the virtual characteristics of the image and take advantage of the KNN algorithm to reduce the error rate without much affecting the accuracy. The experimental results show that the method can effectively decrease the error rate.