Classical Decision Theory has been widely used in multi-agent systems, but are not representative of decision-making in real conditions. So, we built an architecture based on several cognitive psychological theories (e.g. Simon’s decision theory, Tversky’s model of elimination by aspect, etc.) to take human bounded rationality into account. We adopt an intermediate-level of cognitive modelisation, situated between logical level and physical level: the cognition is viewed as modular. In the CODAGE model, the decision maker is modeled by a multi-agent system, where each agent represents a particular sub-process of the whole decision. This framework permits the implementation of cognitive heuristics leading to biaised decision. We illustrate this approach with a simulation of a small experimental financial market, for which our model was able to replicate some human decision behaviors.