Rungie et al. (2005) recently proposed a model that describes the reliability and stability of responses to attitude questions in brand image measurement. We test the validity of this model compared to the model proposed originally by Dall’Olmo Riley et al. (1997) using a new data set which was collected in view of findings by Dolnicar and Heindler (2004) that respondent fatigue has major negative effects on brand image stability. We propose an extension to the proposed model in which we account for heterogeneity in stability across brand-attribute associations. The extended model performs better than the two benchmark models and appears to discriminate well between stable and unstable brand-attribute associations.