BACKGROUND Bystander cardiopulmonary resuscitation (CPR) improves survival. The authors attempted to determine whether the rates at which CPR is performed differ when a cardiac arrest is witnessed by someone known or unknown to the victim. METHODS Retrospective observational cohort study of all witnessed nontraumatic cardiac arrests (Utstein) from Boston from 1994 to 1998. Cardiac arrests were excluded if the original record was unavailable or if medical or public safety personnel witnessed the cardiac arrest. The relationship between the provider of CPR and the victim was determined by the emergency medical technicians at the scene and later categorized as known or unknown. Survival (survival to hospital discharge) was determined through telephone follow-up with the arrest victim's caregivers. RESULTS Known bystanders performed CPR 15.5% (42 of 271) of the time (95% confidence interval (95% CI], 11.2%, 19.8%). Unknown bystanders performed CPR 45.8% (66 of 144) (95% CI, 37.6%, 54.1%) of the time. The odds ratio of receiving CPR if an unknown bystander witnessed a cardiac arrest was 4.61 (95% CI, 2.89, 7.34). Arrests witnessed by unknown bystanders had a 24.3% (35 of 144) (95% CI, 17.2%, 31.4%) survival rate. Those witnessed by known bystanders had a 17.7% (95% CI, 13.1%, 22.3%) survival rate (p = 0.110). In a logistic regression model including both bystander status and location of arrest, unknown bystander status remained statistically significantly associated with having CPR performed regardless of location (OR = 3.56, p = 0.01; 95% CI, 1.64, 7.72). Location was not statistically significant in the presence of bystander status (OR = 1.17, p = 0.686). CONCLUSION Victims of cardiac arrest are more likely to receive CPR when the event is witnessed by bystanders unknown to the victim than if the arrest is witnessed by friends or family.