A rather common way of formalizing contexts as first class objects starts from the basic relation ist(c,p) which asserts that the proposition p is true in the context c. However, the space in which terms take values may itself be context-sensitive. Our aim is to introduce contexts as abstract mathematical entities in a more general framework which includes context-sensitivity, namely knowledge represented by contextual information systems. Making use of some concepts from the Rough Set Theory we refine two relations: the indiscernibility relation between the objects and the similarity relation between the contexts within a contextual information system. Both relations are illustrated with examples showing how contextual information systems can express in a natural way a very few well known phenomena. Based on these relations we propose a simple strategy for decreasing the ambiguity of contextual information systems.