A probabilistic model for representing the discrete brand choice behavior of individuals. On any choice occasion the individual is assumed to choose the item for which he/she has the highest preference. Over repeated choice occasions preferences are assumed to have a probabilistic component. For the logit model this random component of preference is taken to have the double exponential distribution (i.e., the type I Fisher-Tippett extreme value distribution). The model can be used to predict choice probabilities based on attributes of the items being chosen (Corstjens and Gautschi 1983; Yellott 1977). This model is a substitute for discriminant analysis that in addition provides approximate standard errors of estimated model parameters. It is an alternate to regression analysis when the dependent variable is categorical as opposed to continuous. Unlike the probit model or the nested logit model or the Elimination-By-Aspects model, logit assumes that Luce's Choice Axiom holds, which is sometimes seen as a drawback to the use of this model.