It is now twenty years since Rosch and Mervis first published the mass
of evidence on which the Prototype Theory of concepts was originally
based (Rosch, 1977; Rosch & Mervis, 1975). The theory has evolved many
varieties over the years - varieties which have rarely been made expl
icit. These different ways of interpreting the notion of a prototype h
ave often been a worrying source of vagueness and confusion in the the
ory. In this article these different interpretations will be examined
by looking at the kinds of attribute which could be involved in a prot
otype representation, and discussing how the model could be formulated
in each case. It will be argued that a key element required for a suc
cessful model of prototypes is the element of abstraction, and that ce
rtain versions of Prototype Theory that lack this element are inadequa
te as a result.