Two experiments are described in which different groups of participant
s saw the same examples in different orders and then were given an old
-new recognition test. The learning and test examples were created fro
m different combinations from four binary-valued dimensions. One order
(small change) was constructed to maximize the similarity between suc
cessive examples, and the other order (large change) minimized the sim
ilarity across successive examples. The small change condition was con
sistently associated with better old-new recognition than the large ch
ange condition was. These results are discussed in terms of exemplar-g
uided encoding and models of category generalization.