Many aspect mining techniques have been proposed for object-oriented systems. Unfortunately, aspect mining for multi-agent systems is an unexplored research area. The inherent specificities of multi-agent systems (such as autonomy, pro-activity, reactivity, and adaptability) make it difficult to understand, reuse and maintain their code. We propose, in this paper, a (semi-automatic) hybrid aspect mining approach for agent-oriented code. The technique is based on both static and dynamic analyzes. The main motivations of this work are (1) identifying cross-cutting concerns in existing agent-oriented code, and (2) making them explicitly available to software engineers involved in the evolution of agent-oriented code in order to facilitate its refactoring and, consequently, to improve its understandability, reusability and maintainability. The proposed approach is supported by a software tool, called MAMIT (MAS Aspect-MIning Tool), that we developed. The approach and the associated tool are illustrated using a concrete case study.