Confidentiality of certain parameters is an essential security requirement for many security sensitive applications. In this paper, conditions for abstractions are formulated in terms of formal language theory to be able to prove parameter confidentiality in an abstract view of a system and then conclude that an adequate representation of the property is satisfied in the refined system as well. These conditions essentially depend on an agents view as well as on an agent's initial knowledge of the system behaviour, which explicitely formalizes assumptions about the system.
