In the last years, research on agent-oriented (AO) methodologies and multi-agent system (MAS) infrastructures has developed along two opposite paths: while AO methodologies have essentially undergone a top-down evolution pushed by contributions from heterogeneous fields like human sciences, MAS infrastructures have mostly followed a bottom-up path growing from existing and widespread (typically object-oriented) technologies. This dichotomy has produced a conceptual gap between the proposed AO methodologies and the agent infrastructures actually available, as well as a technical gap in the MAS engineering practice, where methodologies are often built <em>ad hoc</em> out of MAS infrastructures, languages and tools.

This paper proposes a new method for filling the gap between methodologies and infrastructures based on the definition and study of the meta-models of both AO methodologies and MAS infrastructures. By allowing structural representation of abstractions to be captured along with their mutual relations, meta-models make it possible to map design-time abstractions from AO methodologies upon run-time abstractions from MAS technologies, thus promoting a more coherent and effective practice in MAS engineering. In order to validate our method, we take an AO methodology &mdash; SODA &mdash; and show how it can be mapped upon three different MAS infrastructures using meta-models as mapping guidelines.