The authors report an ethnographic study of adverse events in medical care in two intensive care units and one general surgical unit at an academic medical center in the United States. Ethnographers attended daily physician rounds, nurse change-of-shift reports, and relevant clinical conferences. The ethnographers used these observations to prospectively detect adverse events in the hospital, regardless of whether they were subsequently recorded in the medical record or discussed with the patient. Adverse events were detected for 46% of studied patients, with 18% classified as severe, with longer length of stay and greater severity of illness as risk factors. The authors suggest that their result, much higher than prior estimates, is more accurate than previous studies because incompleteness and inaccuracy in medical records limit studies based on retrospective chart review. The authors emphasize that 25% of detected errors had “interactive” or “administrative” causes, rather than traditional professional or medical causes, and they recommend increased attention to these sources of error.