[Purpose/significance] Different from the rapid development of the Internet, the development of personal information security protection is relatively lagging. By predicting the gender of social media users, it can better provide privacy protection for the users. [Method/process] The short texts posted by users in social media, Sina Weibo, were taken as the research object. The experiment extracted linguistic features and topic features from the short texts. For each user, we constructed features vector based on linguistic features, topic features, and the superposition of two features, then used SVM Machine learning algorithms built a classifier for gender prediction. [Result/conclusion] Experiments show that the linguistic features and topic features can predict the gender of the users accurately, and the effect is superior to other features used in gender prediction.