We develop a methodology addressing the issue of confounded beliefs and preferences in models of discrete choice. First, we formalize the theoretical framework and logical underpinnings of a belief-preference model of choice for experience and credence goods, where subjective beliefs relate to uncertain product quality. Then, we present the experimental procedure within the context of an online choice experiment studying consumer food preferences. The empirical strategy leverages information from a quality sorting task to identify and estimate beliefs, while choice data are used to recover preferences. By conditioning product choices on predicted quality perceptions, the issue of endogenous beliefs is resolved.