Current biosurveillance systems use free-text chief complaints to determine whether a patient coming to an emergency department has an acute respiratory syndrome that may be due to an infectious or bioterroristic outbreak. We compared our ability to diagnose respiratory syndrome from chief complaints against our ability to detect the syndrome from information contained in dictated emergency department reports. We will describe our results using manually encoded findings processed by a machine learning application called Rule Learner (RL). We will also present initial results on automatically extracting the relevant findings from emergency department reports using pre-existing NLP tools available at the National Library of Medicine.