OBJECTIVE: To determine whether there are disparities in postacute stroke rehabilitation based on type of stroke, race/ethnicity, sex/gender, age, socioeconomic status, geographic region, or service area referral patterns in a large integrated health system with multiple levels of care. DESIGN: Cohort study tracking rehabilitation services for 365 days after acute hospitalization for a first stroke. SETTING: The Northern California Kaiser Permanente Health System (approximately 3.3 million membership population) PARTICIPANTS: A total of 11,119 patients hospitalized for acute stroke from 1996 to 2003. The cohort includes patients discharged from acute care after a stroke. Postacute care rehabilitation services were evaluated according to the level of care ever-received within the 365 days after discharge from acute care, including inpatient rehabilitation hospital (IRH), skilled nursing facility (SNF), home health and outpatient, or no rehabilitation services. INTERVENTIONS: Not applicable. MAIN OUTCOME MEASURE: Service delivery. RESULTS: Patients discharged to an IRH had longer lengths of stay in acute care. Patients with hemorrhagic stroke were less likely to be treated in an IRH. Patients whose highest level of rehabilitation was SNF were older and more likely to be women. After adjusting for age and other covariates, women were less likely to go to an IRH than men. Asian and black patients were more likely than white patients to be treated in an IRH or SNF. Also more likely to go to an IRH were patients from higher socioeconomic groups, from urban areas, and from geographic areas close to the regional rehabilitation hospital. CONCLUSIONS: These results suggest variation in care delivery and extent of postacute care based on differences in patient demographics and geographic factors. Results also varied over time. Some minority populations in this cohort appeared to be more likely to receive IRH care, possibly because of disease severity, family support systems, cultural factors, or differences in referral patterns.