Klin et al., 2004 and Levine et al., 2000 concluded that readers fail to resolve noun phrase anaphors when the antecedent is difficult to retrieve from memory and the inference is not necessary for comprehension. In four experiments we investigated the hypothesis that these inferences were actually partially encoded. Although the results of a lexical decision task demonstrated that readers did not encode a specific lexical item, the results of a reading time measure indicated that they treated the anaphoric noun phrases as co-referential, having reinstated some features from the antecedent episode. We conclude that readers were satisfied with an underspecified representation (Ferreira et al., 2002 and Sanford, 2002); although they knew that an antecedent was present in the passage, they did not devote the resources to fully reinstate it. Further, readers can quickly evaluate the importance of text inputs to comprehension and adjust their attentional resources accordingly.