There exist many approaches to specify and to define security policies. We present here a framework in which the basic components of security policies can be expressed, and we identify their role in the description of a policy, of a system and of a secure system. In this setting, we formally describe two approaches to define policies, and we relate them: the rule-based approach consists of specifying the conditions under which an action is granted and, the property-based approach consists of specifying the security properties the policy aims to enforce. We also show how a policy can be applied to constrain an existing system, and how a secure system can be defined from a security policy.