Two experiments investigated the interpretation of anaphoric noun phrases, and in particular those thatcould only be linked to their antecedents via knowledge-based inferences. The first experiment showed that much of the inferential processing was carried out as the anaphoric noun phrase was read, although there was some indication that inferential processing continued to the end of the clause. The second experiment attempted to establish why anaphoric noun phrases that are more specific than their antecedents cause problems. It showed that the difficulty did not lie in adding the extra information carried by the anaphor to the representation of the referent. Rather, we suggest, putting extra information in the anaphoric noun phrase disrupts the process of linking that noun phrase to its antecedent.