This paper deals with the evaluation of multiagent systems. The related works in this issue are most often interested in evaluation of agent design methodologies or agent tools and platforms but do not address sufficiently evaluation of multiagent applications from a general point of view, the used evaluation criteria remain either specific to a special topic or application dependent. In this work we present an evaluation approach to measure general characteristics of multiagent systems. The proposed method follows a three step process: observation, modelling and measure. The modelling technique is based on graph theory which is used to define and estimate metrics for the evaluation of structural properties of multiagent systems. In this paper, we focus especially in organization as evaluation criterion as it is a very interesting characteristic transposing the social dimension of multiagent systems. Therefore, specific metrics based on graph theory are proposed to analyse agent organizational structures. The tests and experimentations are carried out on a multiagent application for the simulation of production planning and control in supply chains.