OBJECTIVE: Mortality and length of stay are two outcome variables commonly used as benchmarks in rating the performance of medical centers. Acceptance of transfer patients has been shown to affect both outcomes and the costs of health care. Our objective was to compare observed and predicted lengths of stay, observed and predicted mortality, and resource consumption between patients directly admitted and those transferred to the intensive care unit (ICU) of a large academic medical center. DESIGN: Observational cohort study. SETTING: Mixed medical/surgical ICU of a university hospital. PATIENTS: A total of 4,569 consecutive patients admitted to a tertiary care ICU from April 1, 1997, to March 30, 2000. INTERVENTIONS: None. MEASUREMENTS: Acute Physiology and Chronic Health Evaluation (APACHE) III score, actual and predicted ICU and hospital lengths of stay, actual and predicted ICU and hospital mortality, and costs per admission. MAIN RESULTS: Crude comparison of directly admitted and transfer patients revealed that transfer patients had significantly higher APACHE III scores (mean, 60.5 vs. 49.7, p < .001), ICU mortality (14% vs. 8%, p < .001), and hospital mortality (22% vs. 14%, p < .001). Transfer patients also had longer ICU lengths of stay (mean, 6.0 vs. 3.8 days, p < .001) and hospital lengths of stay (mean, 20 vs. 15.9 days, p < .001). Stratified by disease severity using the APACHE III model, there was no difference in either ICU or hospital mortality between the two populations. However, in the transfer group with the lowest predicted mortality of 0-20%, ICU and hospital lengths of stay were significantly higher. In crude cost analysis, transfer patients’ costs were $9,600 higher per ICU admission compared with nontransfer patients (95% confidence interval, $6,000-$13,400). Risk stratification revealed that the higher per-patient cost was entirely confined to the transfer patients with the lowest predicted mortality. CONCLUSIONS: Patients transferred to a tertiary care ICU are generally more severely ill and consume more resources. However, they have similar adjusted mortality outcomes when compared with directly admitted patients. The difference in resource consumption is mainly attributable to the group of patients in the lowest predicted risk bracket.