The computer systems that provide the information under-pinnings for critical infrastructure applications, both military and civilian, are essential to the operation of those applications. Failure of the information systems can cause a major loss of service, and so their dependability is a major concern. Current facets of dependability, such as reliability and availability, do not address the needs of critical information systems adequately because they do not include the notion of degraded service as an explicit requirement. What is needed is a precise notion of what forms of degraded service are acceptable to users, under what circumstances each form is most useful, and the fraction of time such degraded service levels are acceptable. This concept is termed survivability. In this paper, we present the basis for a rigorous definition of survivability and an example of its use.