Donkey pronouns seem to defy the conventional categories of referential and anaphoric pronouns and hence cannot be analyzed as variables. An orthodox treatment is that donkey pronouns are semantically equivalent to definite descriptions. I argue on the contrary that donkey pronouns can be analyzed as bound variables given a distinct notion of binding. I provide a systematic comparison between the static, description-theoretic approach and Discourse Representation Theory (DRT), a representative of dynamic semantics. By exposing the inadequacy of various descriptivist theories, I motivate and argue that DRT is the better alternative. DRT is superior for being a coherent and flexible analysis of donkey pronouns, a unified analysis of pronouns in general, and an intuitively appealing model of meaning. In addition, I uncover the similarities between the situational descriptivist account and DRT. I show that when fully elaborated, the former turns out to be a notational variant of the latter. I then trace their common problems to the Lewisian assumptions of quantification and conditionals; my proposed solutions suggest non-trivial modifications to and clarifications of the underlying Lewisian framework.