I analyze a corpus of 600 examples of backwards anaphora collected from written English texts, limited to examples of genuine backwards anaphora where the full noun phrase is clearly the antecedent for the pronoun. I show that certain constructional configurations recur with overwhelming frequency, while other configurations occur rarely or not at all (though they do occur as examples of Ôrepeat identificationÕ or re-naming of a referent that is previously established in the discourse). The patterns are explained by the requirement that genuine backwards anaphora involve a marked prominence asymmetry between the pronoun and antecedent, where prominence is defined in the conceptual-semantic terms of the theory of Cognitive Grammar. I offer some preliminary observations on the discourse functions of backwards anaphora.