Four studies show that consumers have not one but two distinct learning pro
cesses that allow them to use brand names and other product features to pre
dict consumption benefits. The first learning process is a relatively unfoc
used process in which all stimulus elements get cross-referenced for later
retrieval. This process is backward looking and consistent with human assoc
iative memory (HAM) models. The second learning process requires that a ben
efit be the focus of prediction during learning. It assumes feature-benefit
associations change only to the extent that the expected performance of th
e product does not match the experienced performance of the product. This p
rocess is forward looking and consistent with adaptive network models. The
importance of this two-process theory is most apparent when a product has m
ultiple features. During HAM learning, each feature-benefit association wil
l develop independently. During adaptive learning, features will compete to
predict benefits and, thus, feature-benefit associations will develop inte
rdependently. We find adaptive learning of feature-benefit associations whe
n consumers are motivated to learn to predict a benefit (e.g., because it i
s perceived to have hedonic relevance) but find HAM learning when consumers
attend to an associate of lesser motivational significance.