We derive a family of probabilistic choice models including the multinomial logit model, from a microeconomic model in which the decision maker has to make some effort in order to avoid mistakes when implementing any desired outcome. The disutility of this effort enters the decision maker's goal function in an additively separable way. A particular disutility function, yielding the multinomial logit and GEV models as special cases, is characterized axiomatically. Unlike the usual random-utility approach, the present approach leads to a normalization of the achieved utility with respect to the number of alternatives. The present model also applies to continuum choice sets in Euclidean spaces, and provides a microeconomic foundation for quantal response models in game theory.