The authors develop a class of mixtures of piece-wise exponential hazard models for the analysis of brand switching behavior. The models enable the effects of marketing variables to change nonproportionally over time and can, simultaneously, be used to identify segments among which switching and repeat buying behavior differ. Several forms of asymmetry in brand switching are accommodated. The authors provide an application to the analysis of scanner panel data on ketchup, which illustrates the implications for asymmetry, nonproportionality, and heterogeneity. The results show that the model predicts purchases and purchase timing in holdout data better than the models proposed by Kamakura and Russell (1989) and Vilcassim and Jain (1991).