In a mixed-choice situation, consumers must choose between brands that are directly observable and brands encountered previously, Prior research suggests a conservative bias to choose an observable brand if it is acceptable and if memory for the previously encountered brands is poor. The authors demonstrate in three experiments that this bias may be highly sensitive to stimulus factors. They replicate the bias when all brands have similar attribute structures. However, when the brands are less comparable, aspects of the previously encountered brands or their descriptions may interact with memory to eliminate or reverse the effect. Thus, over time, preference for a memory brand in relation to an observable brand may increase or decrease significantly, and may do so irrespective of the memory brand's true relative merits. Implications of these results for current models of persuasion ore discussed.