This paper describes a way to make social theories of privacy useful and relevant for policy guidance. Privacy as a human right draws on the well-developed rhetoric and style of argumentation of the international human rights movement. Privacy as harm prevention plausibly draws on the utilitarian tradition of assessing the costs and benefits of privacy protections versus their economic costs. A social theory of privacy cannot draw on these familiar policy frameworks. This paper attempts to remedy that defect. It first examines theories that treat privacy as an element of social structure including the views of Nissenbaum, Merton, Foucault, Bloustein, Post, Regan, Steeves, and Benkler. I develop a notion of harm to a social context which arises when individuals avoid engaging in certain social practices. I examine the contextual harm basis for several witness privileges: priest-penitent, attorney-client, psychotherapist-patient, reporter-source, and doctor-patient. I apply the notion of contextual harm to genetic privacy, student privacy, and to privacy in online social networks. I recommend a policy principle that privacy restrictions should be considered whenever an information flow causes or is likely to cause significant contextual harm. I contrast this principle with familiar privacy principles such as purpose specification and respect for context. I also seek to understand how to resolve cross-contextual conflicts where gains in one context mean losses in others, an ever-more urgent task in light of big data’s capacity to aggregate and decontextualize information. While the principle of contextual harm is useful in providing a basis for many privacy restrictions, it is not complete. An information flow that causes no contextual harm might still be objectionable. Other normative bases for privacy restrictions, drawn from the considerations of human rights and utilitarianism, will need to be invoked for a full privacy theory.