The ability to detect the students’ motivational state during an instructional interaction can bring many benefits to the performance of an Intelligent Tutoring System (ITS). In this paper we present an empirical study which provided us with a considerable amount of knowledge regarding motivation diagnosis. We show how this knowledge was formalised in order to create a set of motivation diagnosis rules that can be incorporated into a prototype tutoring system. We also briefly present how these motivation diagnosis rules were evaluated in another study.