In discourse processing, speakers collaborate toward a shared mental model by establishing and recruiting prominence relations between different discourse referents. In this article we investigate to what extent the possibility to infer a referent's existence from preceding context (as indicated by the referent's information status as inferred or brand-new) and a referent's unique identifiability (as indicated by a referent's uniqueness status) affect (1) ambiguous pronoun resolution in comprehension and (2) the bias to mention a referent again and make it topic in discourse planning. In Experiment 1, a visual-world eye-tracking experiment, we found that ambiguous pronouns are more likely to be interpreted as linked to the direct object of the preceding sentence when the associated referent was inferred and unique than when it was inferred and non-unique. For brand-new referents, uniqueness status did not affect ambiguous pronoun resolution. In Experiment 2, a story continuation experiment, we found that inferred and unique referents were mentioned again and made topic less often than inferred and non-unique referents as well as brand-new referents. Results are discussed within a dual-process activation model, which distinguishes the activation of a noun phrase's concept through inference relations and the activation of a noun phrase's referent through the referent's uniqueness status.