Social media’s role in facilitating anti-authoritarian protests has received much recent attention. Although a handful of regimes like Tunisia and Ukraine have undergone major changes, savvy autocrats elsewhere have co-opted online space with propaganda while censoring to prevent opposition. Yet in China and other cases, we sometimes observe less censorship than conventional wisdom about authoritarian information control would predict. Why do some autocrats choose to censor selectively, and how do they actually implement such fine-grained control? In this project, I argue that allowing limited online criticism can signal regime responsiveness to public demands on issues where leaders' legitimacy is at stake. I develop this logic through a focus on China. Chinese Internet industry interviews address the why and how -- i.e. the elite beliefs, and bureaucratic apparatus -- behind China’s selective censorship since 2011. Second, social media data analysis of online incidents on Sina Weibo (China's Twitter) reveals that censorship is selective even within sensitive issues. The implication of these findings is that leaders' ability and willingness to fine-tune censorship may be vital to maintaining popular support (or forestalling dissent) among increasingly educated, urban, Internet-literate publics whose views are crucial to regime survival in rapidly developing authoritarian states.