Consumers use brand names and product features to predict the performance o
f products. Various learning models offer hypotheses about the source of th
ese predictive associations. Spreading-activation models hypothesize that c
ues acquire predictive value as a consequence of being present during the a
cquisition of product performance information. Least mean squares connectio
nist models hypothesize that any one cue acquires predictive value only to
the extent that it can predict differences in performance that are not alre
ady predicted by other available cues. Five studies in the context of portf
olio-branding strategies provide evidence supporting a least mean squares c
onnectionist model. As predicted by this model, results show that subbrandi
ng and ingredient-branding strategies can protect brands from dilution in s
ome situations but can promote dilution in other situations.