Weibo allows users to add text tags in their profiles, which are descriptive to one's personality and interests. The tag information can be very useful to user profiling in applications such as personalized recommendation, expert finding and social influence measuring. This paper first studies the characteristics of users' tagging behavior and content of the tags based on large-scale data. By adopting topic model on users' Weibo posts, it finds that the more tags two users have in common, the more similar their Weibo posts are and vice versa. It also finds that the users with connections to each other have more similar tags and Weibo posts. Based on this observation, this study uses tags and Weibo posts to predict user connections separately on real-world data. The experimental results show that the tag-based approach is significantly better than the approach based on Weibo posts, thus validating the effectiveness of user tags in describing user interests.