Accepted by........

As the Internet has grown in popularity, so have efforts to control and monitor its use. Countries such as China [46] and Saudi Arabia [25] already block certain kinds of content and more may follow as censorship tools become increasingly sophisticated. In response, Feamster et al. have proposed Infranet [10], a system allowing clients to retrieve sensitive content through a distributed network of secretly cooperating web servers. Infranet differs from other anti-censorship system in that it surreptitiously embeds requests and responses for sensitive content inside what appear to be innocuous web transactions. It is the first anti-censorship system that attempts to achieve covertness in addition to confidentiality. This thesis presents several challenges that we encountered while implementing the first public release of Infranet. Our research has yielded a variety of innovative techniques to improve the speed, security, and scalability of Infranet. We describe a new technique for generating probabilitistic models of client behavior, as well as algorithms that leverage this knowledge to provide more efficient communication between clients and servers. We explore several attacks a censor might mount an