In this paper, we study the problem of detecting rumors spreading in the social networks. Different from the most of the previous works on identifying rumors in Twitter, we select Sina Weibo, the China's major microblog system, as our target. We use two interfaces named "@Weibopiyao" and "Weibo Misinformation-Declaration" from Sina Weibo to help us construct high accuracy training dataset. We analyze data types of microblogs based on their content and the role and possible social impacts of different types of microblogs in rumors spreading. Leveraging our findings, we then focus on detecting social news rumors on Weibo. A new method is proposed to annotate the collected data from Weibo automatically, and three new features for identifying social news rumors are proposed. Experimental results illustrate the efficacy and efficiency of the methods and features proposed in this paper.