Previous approaches to information modelling (IM) use as the first modelling step an identification of the classes or types of things needed to describe the domain. The paper shows that this assumption is inappropriate in many applications of IM.