Abstract : XML technology and multi-agents systems are new approaches offering interesting assets for corporate memory management. Since ontologies appear as key assets in the new generation of information systems and also in the communication of multi-agents systems, it comes with no surprise that it stands out as a keystone of multi-agents information systems. We present here the overall approach of the CoMMA project, and then focus on the first elements of our return on experience in building the ontology O " CoMMA used by this multi-agent corporate memory system. With our entrance in the information society and the associated shift in economical rules of game, organizations had to adapt and their interest in corporate knowledge capitalization and management grew stronger. The semantic web technologies provide interesting techniques to materialize and structure memories to prepare their exploitation and management. At the same time, distributed artificial intelligence proposes appropriate paradigms, especially the multi-agents one, to deploy a software architecture over this distributed information landscape. Through intelligent collaboration agents can achieve a global capitalization of the corporate knowledge while being able to locally adapt to individual resources and users specificity. In a first part we present the overall approach of CoMMA, and we then focus on the methodology and our return on experience in building an ontology for such a system.