Abstract Objective: To assess the relation between the source of stroke patients' referral to the rehabilitation unit and rehabilitation outcome. Design: Multivariate analysis of rehabilitation outcomes prospectively collected and supplemented by data from Department of Veterans Affairs (VA) administrative records. Setting: Fifty-three hospital-based VA rehabilitation units located nationwide. Patients: A total of 2,031 first-admission stroke patients discharged from VA rehabilitation units in fiscal years 1993–1994. Interventions: Observational study. Outcome Measures: Functional Independence Measure (FIM) at admission, gains in motor and cognitive function, rehabilitation length of stay (LOS), and the probability of discharge to home. Results: Controlling for clinical, demographic, and unit characteristics, referral source predicted admission function, LOS, and home discharge ( p ≤ .0001), but did not predict motor or cognitive gain ( p > .05). Patients referred from long-term care and other acute hospitals (Acute Other) had significantly lower adjusted admission FIM scores than patients referred directly from home (respectively, 12.9 and 17.0 FIM units lower), and from the acute care services of the VA hospital providing rehabilitation (4.8 and 8.9 units lower). Acute Other referrals had longer risk-adjusted LOS than referrals from any other source (4.6 to 6.4 days longer), and patients from both Acute Other and long-term care had a lower probability of home discharge than the remaining two referral groups (reduction in probability range: .06 to .27). Conclusion: VA administrators and clinicians need to consider referral source when evaluating and comparing unit performance in stroke rehabilitation. Patients referred from Acute Other and long-term care settings may require more intensive management to offset their functional and social deficits, and to reduce LOS and increase home discharge rates. Referral source also identified highly functional patients who possibly could be cared for appropriately in less costly rehabilitation settings. These issues warrant further research especially as the VA seeks to achieve cost savings in this era of reduced funding.