This case-control study of 31 specialized dementia units and 32 traditional units in five states investigated use of physical and pharmacologic restraints among 625 patients with the diagnosis of dementia. Physical restraints were observed in use on 18.1% of dementia unit patients and on 51.6% of comparison unit patients who were out of bed during the day (adjusted odds ratio, 0.283;95% confidence interval, 0.129 to 0.619). Pharmacologic restraints were routinely given to 45.3% of dementia unit patients and 43.4% of comparison unit patients (adjusted odds ratio, 0.950; 95% confidence interval, 0.611 to 1.477). We used multivariate logistic regression to identify residence in a nonspecialized nursing home unit, nonambulatory status, transfer dependency, mental status impairment, hip fracture history, and a high nursing staff-to-patient ratio, which we found to be independent predictors of physical restraint use. Physically abusive behavior, severe mental status impairment, and frequent family visitation were found to be significant predictors of pharmacologic restraint use, while advanced patient age, large nursing home size, and patient nonambulatory status were protective against such use. These results support the conclusion that physical and pharmacologic restraint constitute separate treatment modalities with different risk factors for use, and indicate that specialized dementia units are successful in reducing the use of physical but not pharmacologic restraints.