Developing context-aware applications needs facilities for recognizing and representing context, reasoning on it and adapting to it accordingly. In what concerns context representation, the newest and most challenging representation is the ontological one. The problem is that current ontologies for context are quite simplistic and do not provide a standard for representing complex context attributes. In this paper, we propose a context definition and representation used to construct a context-based agent architecture. The representation we propose combines the generality provided by ontologies with the complexity inspired by the object oriented models. The goal of the proposed representation is to support the deployment of context-aware agents able to learn how to recognize and adapt to the context of their decisions.