In conventional random-utility models, such as the multinomial logit model, it is assumed that a decisionmaker's choice set is independent of his or her preferences conditional on the explanatory variables of the models. However, there are many situations in which the decisionmaker chooses the choice set, thereby making the independence assumption implausible. For example, if information about alternatives is costly, an individual may choose to learn in detail about only a small group of them. This paper is concerned with the modeling of discrete choice with endogenous choice sets. It is assumed that costly information forces the decisionmaker to acquire detailed knowledge of only a small group of alternatives that she or he selects. Models are developed for choice-set generation in this context and for choice conditional on endogenous choice sets.