Assessment of neuropsychiatric symptoms in dementia has great clinical importance. The aim of the study was validation of the Danish version of the NPI, using assessments of 72 demented and 29 non-demented of age 65+ years and their caregivers at three visits. The NPI was administered by the same psychiatric nurse interviewing the same caregiver. At visits 1 and 3, a psychogeriatrician assessed the participant using the ICD-10, the Geriatric Deterioration Scale (GDS) and the Clinical Global Impression (CGI) as well as the NPI in a visual analogue scale (VAS) version. These scores were blindly converted into scores equalling the frequency and severity of the NPI by one of the investigators. Data analysis comprised inter-rater reliabilities (intra-class coefficients, ICC); NPI scores and corresponding VAS scores were compared using Spearman's correlation coefficients. NPI scores at visits 1 and 2 were used to assess the test–retest reliabilities. The scalability of the NPI was assessed with Mokken and Loevinger coefficients. The ICC for all the NPI domains and the GDS (>0.80) were perfect, the ICC for the NPI-VAS (0.68–0.95) and the CGI (0.69) was satisfactory to perfect. Correlations between NPI and NPI-VAS were high; only two domains had coefficients below 0.60: depression and agitation/aggression. NPI-total scores increase with increasing severity of dementia. The NPI did not fulfil the scalability assessed by the Mokken and Loevinger coefficients. The NPI Danish version is valid and reliable in assessing neuropsychiatric symptoms in dementia but not fully scalable. The use of single item scores and not total sum score is recommended.