High quality image data on how consumers perceive brands is essential to make good brand management decisions. Prior studies reveal that brand images are not very reliable, as they are typically measured in industry, which might be due to the answer format typically used (Rungie et al., 2005). The practical implication is that brand image data — as currently collected in consumer surveys — is not a valid source of market information. We challenge this implication. Using three measures of stability we test whether the binary answer format produces image data less reliable than alternative formats. We investigate whether the aggregate descriptive model of brand image stability proposed by Rungie et al. can be improved by accounting for heterogeneity. Results indicate that, compared to alternative formats, binary answer formats lead to equal stability levels, and most brand-attribute associations are stable. Unstable associations typically fail to describe adequately the brands under study. Practical implications include that binary brand-attribute associations can be used safely to measure brand images. Also, practitioners can get guidance about required brand management measures by discriminating between stable and unstable brand-attribute associations. A model that helps managers classify brand-attribute associations into stable or unstable is proposed in the article.