Three experiments are reported which deal with the nature of categoriz
ation of natural objects. Stimulus materials were schematic butterfly
patterns. Systematic transformations were given to the original patter
n to make stimulus instance. In Experiment 1, subjects were given a re
cognition memory task with recognition ratings. Recognition ratings we
re found to be related to family resemblance score (FRS: the sum of fe
ature frequency in the category) rather than to transformation or subj
ective similarity to the original. In Experiment 2, subjects were give
n the recognition memory task while the FRS was kept constant. Weight
of each feature affected recognition ratings differently, when FRS was
kept constant. In Experiment 3, subjects were given the same task as
in Experiment 2, while the FRS and the component frequency of each fea
ture were kept constant. recognition ratings differently. These findin
gs suggest that both the frequency and weight of feature itself affect
recognition ratings.