Consumers often do not account for unplanned purchases in their store choice when planning grocery trips. I develop a discrete choice model with partially unobserved attributes to capture such a behavior. The model is partially identified, and the sharp identified set is characterized via likelihood-based criteria. The inference approach generates the pseudo identified set even in the presence of model misspecification. Using household grocery shopping data, I show that point estimation from the standard multinomial choice model assuming no unplanned purchases is rejected; I instead obtain confidence set for consumers’ preferences.