Using an experiment built into a longitudinal survey, we demonstrate that the low stability of consumers' brand-attribute associations [see Rungie, C., Laurent, G., Dall'Olmo Riley, F., Morrison, D.G., & Roy, T. (2005). Measuring and modeling the (limited) reliability of free choice attitude questions. International Journal of Research in Marketing, 22(3), 309–318.] can be partly explained by the prevailing methods used in market research, which can often lead consumers to construct temporary associations. To increase the proportion of stable brand-attribute associations, we recommend the following improvements in market research methods: use of a shorter, brand-prompted attribute association task; inclusion of an “don't know this brand” option to isolate ratings of brands unknown to the individual; omission of the standard instruction to guess when uncertain; and, in cross-cultural studies, translation of instructions and attribute descriptions into appropriate first languages. Even with these improvements, however, the maximal stability of associations that brands can achieve is less than 100% after correcting for methodological influences. This imperfect stability may mean that consumers learn brand-specific attribute associations that are temporary but stable enough to lead them to try or re-try the brand, after which consumers replace the specific brand associations with a summary brand attitude.