This paper evaluates the classification support provided by current object-oriented analysis methods and proposes possible improvements. Concepts of object-orientation are briefly reviewed. Key principles for classification are identified and emerging object-oriented analysis methods are then judged against these principles. In an attempt to avoid some of the identified shortcomings, it is proposed that classification support during analysis should be based on two distinct constructs (types and collections) instead of a single class construct. This suggestion is based on a clear distinction between the essential and the contingent properties of the modelled objects.