OBJECTIVES: To observe the incidence of problems associated with use of medical alarms in intensive care units. METHODS: Adverse medical device events (AMDE) and potential patient harm (hazards) associated with alarms in two intensive care units (Shock-Trauma, STICU, and Medical-Surgical, MSICU) within a tertiary referral hospital were observed by a intensive care nurse native to each unit and trained in observational techniques by an anthropologist. The study was conducted during consecutive 6-month periods between October 2001 and September 2002. Five rooms were continuously directly monitored during each of approximately 104 six-hour shifts (4 randomly selected shifts per week). Poisson-method 95% confidence intervals around incidence rates were calculated. RESULTS: A total of 81 alarm events were observed in the monitored rooms; 40 were AMDEs and 41 hazards. Forty-five events were related to respiration, oxygenation, or ventilation monitoring, 22 to blood pressure, heart rate, or arrhythmia monitoring, 9 to an infusion pump, and 1 to a feeding pump. Incidence rates per 100 occupied observed bed-days for hazards and AMDE (with 95% confidence intervals) in the STICU were, respectively, 20 (12-31) and 35 (24-49). The corresponding incidence rates for the MSICU were 20 (13-31) and 6 (2-12). A human factors analysis of the circumstances of each case attributed 79% to a combination of user error and device malfunction or flawed alarm design. None of the events were documented in the medical record or in incident reports. CONCLUSIONS: Rates of alarm related problems in two ICUs were much higher than reported in a routine reporting system and indicate an understudied important threat to patient safety. An assessment of risk factors for these events is in progress.