Purpose: The purpose of this study was to quantify the effect of specific nursing home features and state Medicaid policies on the risk of hospitalization among cognitively impaired nursing home residents. Design and Methods: We used multilevel logistic regression to estimate the odds of hospitalization among long-stay (greater than 90 days) nursing home residents against the odds of remaining in the nursing home over a 5-month period, controlling for covariates at the resident, nursing home, and county level. We stratified analyses by resident diagnosis of dementia. Results: Of 359,474 cognitively impaired residents, 49% had a diagnosis of dementia. Of those, 16% were hospitalized. The probability of hospitalization was negatively associated with the presence of a dementia special care unit (adjusted odds ratio [AOR] = 0.90, 95% confidence interval [CI] = 0.86-0.94) and with a high prevalence of dementia in the nursing home (AOR = 0.96, 95% CI = 0.88-1.03). Higher Medicaid payment rates were associated with reduced likelihood of hospitalization (AOR = 0.95, 95% CI = 0.90-1.00), whereas any bed-hold policy substantially increased that likelihood (AOR = 1.44, 95% CI = 1.12-1.86). We observed similar results for residents without a dementia diagnosis. Implications: Directed management of chronic conditions, as indicated by facilities' investment in special care units, reduces the risk of hospitalization, but the effect of bed-hold policies illustrates how fragmentation in the financing system impedes these efforts. (Contains 3 tables.)