What symptoms are associated with the patient’s genotype? Did patients treated with medication fare better than patients treated surgically? Which patients are more likely to be readmitted to the hospital? Many of the pressing problems in health care today require access to detailed information locked in narrative reports. Natural language processing (NLP) offers access to symptoms, risk factors, diagnoses, and treatment outcomes described in text. Researchers have been applying NLP to clinical text for many decades, but NLP is far from being a mainstream technology in health care. In this talk I will help explain the challenge in developing and applying NLP to clinical data, I will describe the work our research lab has done to improve access to the rich data contained in narrative reports, and I will call for principled informatics approaches when approaching the problem of information extraction from clinical text.