Under the tide of information technology revolution, social network services (SNS) become a typical application in Web2.0 era by its rich and interactive user participation, and has swept the world in a short time. More and more users begin to express themselves on Facebook, Micro-blog and other social networks, and user features are existing in the SNS in a more intuitive way. These information assign complete personality and image for each node in the SNS, which has enormous potential commercial value. For background, this paper studies the user features mining problem in social networks, and focuses on the user features without labels. Firstly, two models are established for user features mining without labels, in which clustering, classification, text mining and graph mining are used in the model. Then, the proposed models are implemented under two scenarios: user interest discovery and Micro-group structure discovery. Experiment results based on Sina Micro-blog show that the accuracy is above 80%, which can meet the demand of user features mining in social networks. Therefore, the techniques proposed in this paper are feasible.