Adaptive behavior modeling requires to cope with cognitive tasks like motivation, action selection or temporal organization of behavior. In this framework, cortical based models have proven to be efficient : the distributed organization and combination of information in monomodal (sensory or motor) and polymodal (associative) areas together with a statistical and slow learning can allow an efficient information processing at several levels of integration. But, while necessary, these cortical models cannot directly handle episodic memory mainly because it requires fast learning abilities as well as strong combinatory properties. Several results from neurobiology show that all these associative cortical areas project on to the entorhinal cortex (EC). It holds at any time an integrated representation of the whole cortical activity which can be thus learned by the adjacent hippocampal structure and possibly re-instantiated within EC: cortex is then offered a way of manipulating episodic memory. To explore that mechanism, we designed a computational model of hippocampal-cortical interaction grounded on a synaptic triad mechanism acting on the EC and controlled by the hippocampus. In our model, each EC unit is connected to all other EC units via synaptic triads (i.e. synapses modulated by other units). Each of these synapses (which may be excitatory or inhibitory) is actually modulated by every active unit of CA1, a sub-structure of the hippocampus. This latter has been designed according to neurobiological data and is composed of three main structures: * Dentate gyrus (DG): It is directly linked to the EC and ensures information dispersion since cortical patterns may overlap greatly. * CA3: It receives information from the DG and performs first, information re-compression, and second, auto-associative memorization using recurrent links present within this structure. * CA1: It receives information from CA3 and compresses it again Once the loop [EC -> DG -> CA3 < -> CA3 -> CA1] has been performed, CA1 holds a representation corresponding to EC activity. It is then able to act on the synapses of the EC. This action will be directly dependent on what has been learned before: each presented pattern is implicitly compared with memorized ones and, depending how close it matches, the presented pattern is learned or a previous one is instantiated. Current experiments and results show that this mechanism allows an efficient memorization of patterns within hippocampus and re-instantiation within EC. It is to be noted that the pattern of connectivity we used between CA1 and EC is induced by the absence of topological properties within the entire model and it consequently does not match available data on hippocampus connectivity. We are thus working on the implementation of topological connections between structures in order to dramatically reduce connection rates. Furthermore, a feedback loop involving the septum is currently studied in order to be able to dynamically modulate learning.