We evaluated the floor effect and convergent, discriminant, and known-group validity of the Bedford Alzheimer Nursing Severity scale (BANS-s), a rating scale comprising cognitive and functional items recently developed for grading severe dementia. Ninety-nine demented patients (81 females and 18 males aged 55-100 years) in two nursing homes were assessed with the BANS-s, established cognitive and functional scales [Mini Mental State Examination, the extended version of the Clinical Dementia Rating (CDR), Katz's basic activities of daily living, Tinetti balance and gait, and Crichton scales], a behavioral scale (UCLA Neuropsychiatric Inventory), and indicators of malnutrition (Prognostic Nutritional Index). A relevant proportion (40%) of patients scored close to the floor of all scales except BANS-s and CDR, which showed a more uniform distribution of scores throughout the possible range. Convergent validity of BANS-s with the other cognitive and functional scales was good, with Pearson's r ranging from 0.62 to 0.79. Discriminant validity analysis of BANS-s versus the UCLA Neuropsychiatric Inventory showed that the two scales measure different domains (Pearson's r = 0.36). To test known-group validity, all patients were divided into two groups of different severity as defined by the Prognostic Nutritional Index. BANS-s and CDR were the scales with the best ability to discriminate malnourished from nonmalnourished patients. As a further validity test, the 37 patients reaching the floor on other cognitive and functional scales were divided into two subgroups of different dementia severity as defined by the Tinetti scale. BANS-s but not CDR was able to differentiate the two groups.