One hundred thirty-five (52.9%) of 255 patients were given antipsychotics at baseline, 40.4% took conventional neuroleptics, and 12.5% took atypical antipsychotics. After 2 years, 118 of 255 were deceased. The death rate was 47.6% among those taking conventional antipsychotics, 21.9% among those taking atypical antipsychotics, and 50.0% among nonusers. We performed a logistic regression analysis to clarify which factors had independent prognostic value in mortality. When we entered age, gender, severe stage of dementia (clinical dementia rating=2–3), delirium, high number of comorbidities, impaired physical functioning, use of neuroleptics, use of atypical antipsychotics, and use of restraints into the model, only old age (>85 years) (odds ratio=1.71, 95% confidence interval [CI]=1.00–2.95), high number of comorbid disorders (odds ratio=1.96, 95% CI=1.03–3.73), and use of restraints (odds ratio=2.45, 95% CI=1.06–5.65) predicted mortality. It is surprising that the use of atypical antipsychotics seemed to protect against death (odds ratio=0.40, 95% CI=0.17–0.96). Conventional neuroleptics did not have any effect.