OBJECTIVE To assess whether vital sign measurements could identify internal medicine patients at risk for cardiopulmonary arrest. METHODS Retrospective case-control study comparing 72 hours of pre-arrest vital sign measurements with 72 hours of vital sign measurements for patients from the same units who did not experience cardiopulmonary arrest. METHODS Twelve non-intensive care internal medicine units at a large midwestern academic medical center. METHODS Cases included all 59 inpatients who had experienced cardiopulmonary arrest between May 1989 and December 1990; patients who were designated do-not-resuscitate (DNR) or had less than 72 hours of vital sign recordings were excluded. Controls included 91 inpatients without cardiopulmonary arrest who were matched for units and who had 72 hours of vital sign recordings. RESULTS The occurrence of one or more respiratory rates > 27 breaths per minute over a 72-hour period had a sensitivity of 0.54 and a specificity of 0.83 (odds ratio = 5.56, 95% CL = 2.67-11.49) in predicting cardiopulmonary arrest. Other respiratory rate thresholds were also predictive of arrest. The ability of respiratory rate to predict arrest was stronger in units with high incidences of arrest relative to units with low incidences, for example, in units for the management of gastrointestinal disease (sensitivity = 1.00, specificity = 0.86) and renal disease (sensitivity = 0.69, specificity = 0.87). Respiratory rate remained a significant predictor (p < 0.001) after controlling for patient age and gender. Pulse rate and blood pressure were not predictive of cardiopulmonary arrest. CONCLUSIONS Using elevated respiratory rates as a signal for focused diagnostic studies and therapeutic interventions in internal medicine patients may be useful in reducing the incidence of subsequent cardiopulmonary arrest, and lowering associated morbidity and mortality.