Although discrete-choice statistical techniques have been used with increasing regularity in demographic analyses, McFadden's conditional logit model is less well known and seldom used. Conditional logit models are appropriate when the choice among alternatives is modeled as a function of the characteristics of the alternatives, rather than (or in addition to) the characteristics of the individual making the choice. We argue that this feature of conditional logit makes it more appropriate for estimating behavioral models. In this article, the conditional logit model is presented and compared with the more familiar multinomial logit model. The difference between the two techniques is illustrated with an analysis of the choice of marital and welfare status by divorced or separated women.