When the costs of dementia are analyzed, interviews regarding caregiver time (both informal or formal) and other kinds of resource use are frequently used. However, it is unclear how valid such data are. The aims were to investigate the validity and reliability of interview-based estimated resource use, particularly caregiver time. Twenty institutionalized demented persons were studied during 3 months. A special team recorded the amount of time the caregivers spent on personal activities of daily living (PADL), instrumental activities of daily living (IADL), and supervision. After each session, the caregivers estimated the amount of time they had spent on care. Observed time was regarded as the golden standard. Intra-rater reliability was tested with proxy informants on 25 elderly living at home. In total, 110 observations and 108 estimations of caregiver time were made. The correlation coefficient between observed and estimated PADL-time was 0.81 (p < 0.001), in IADL 0.29 (p = 0.03), and in supervision 0.51 (p < 0.001). The intra-rater reliability was high for almost all items (Cronbach's alpha and intraclass correlation coefficient >0.9). There was a high correlation between interview-based data and register data regarding hospital care, family physician, and district nurse visits. In conclusion, interviews may serve as a valid and reliable substitute for observations.