This thesis motivates a paradigm shift from keyword search toward allowing people to express their information needs naturally to search engines in the same way they might explain what they were looking for to another person. Natural language search offers two immediate benefits realizable today: improved retrieval accuracy via more accurate query interpretation, and improved usability by making search engine use consistent with human interaction. Natural language also defines the appropriate representation of information needs to focus long-term research in information retrieval (IR) toward matching human-level competence in our automated systems. Supporting natural language, however, presents a significant challenge to existing IR methodology which has primarily focused on supporting keyword search. This thesis provides an early response to this challenge, demonstrating that easier, more accurate search can be realized today while at the same time highlighting the long road before us as we work toward meeting IR's Turing Test.