In the past couple of decades, agent-oriented technology has been arisen in order to assist in developing intelligent software that is able to solve challenging problems. Numerous methodologies for developing agent-based systems have been proposed in the literature and the area of agent-oriented methodologies is maturing rapidly. Evaluating methodologies' strengths, weaknesses and domains of applicability plays an important role in improving them and in developing the "next-generation" of methodologies. In this paper, we present a reliable framework that adopts statistical techniques to compare agent-oriented methodologies. Based upon this framework we performed a comparison of four AOSE methodologies MaSE, Prometheus, Tropos and Gaia.