Four experiments were designed to test two predictions of Prototype Th
eory. The first prediction was that when the defining (necessary) feat
ures of a concept are only partially matched by an instance, then char
acteristic (nonnecessary) features of concepts can affect categorizati
on. The test of this prediction was rendered problematic as successive
experiments failed to identify clearly necessary features for a range
of concepts. The second hypothesis related to the independence of fea
tures in determining similarity. Most versions of Prototype Theory ass
ume a linear combination of feature matches, which would predict that
the effect of changing a feature on category membership should be grea
test when the probability of categorization is closest to 50% (i.e. at
the category border). The results showed that, contrary to this predi
ction, the effect of changing a feature was greatest when other featur
es were all positive, and so categorization probability was at a maxim
um. The results support either a logistic combination rule for assessi
ng similarity on the basis of feature match (Medin & Shaffer, 1978), o
r an exponential generalization function relating similarity to protot
ype to the sum of matching features (Shepard, 1987). (C) 1995 Academic
Press, Inc.