With the rapid development of micro-blog, spam detection and filtering is faced with enormous challenges. It is significant to realize realtime and accurate detection of spam, which is important to improve user experience and the sustainable development of micro-blog platform. In this paper, a spam detection method based on multi-features of micro-blog is proposed. The main procedures are:first, the features of user and content are extracted. Second, LDA is applied to extract latent topic features. Finally, the features above are fused and a proper classifier is trained based on SVM. Experimental results show that the precision and F1 get increased while adopting the method proposed in this paper compared to the pervious methods.