OBJECTIVE To evaluate the usefulness of registered nurses' (RNs) and nursing assistants' (NAs) pain report among 304 institutional older people with dementia and elucidate the factors influencing the agreement of pain reports within dyadic assessors. BACKGROUND Using informant reports has been suggested as an effective strategy to solve the dilemma of pain assessment in people with dementia. However, the validity of pain information provided by institutional caregivers is still under-investigated. METHODS A prospective study was undertaken in six dementia special care units in Northern Taiwan. METHODS Both the four-item pain inventory and Doloplus-2 scale were used to investigate different raters' pain reports. RESULTS Across pain variables, fair to moderate agreement was noted in RNs' assessment (Kappa = 0.29-0.42), but poor agreement in NAs assessment (Kappa = 0.11-0.24). Comparing mean scores on intensity, frequency of pain and Doloplus-2, statistically systematic bias existed in most pairs, but the greater bias occurred in NAs' assessment. In logistic regression analysis, when pain was assessed by the pain inventory, residents' cognitive impairment level and caregiver's age were associated with the agreement of pain presence perceived in either resident-nurse or resident-NA pairs. Nevertheless, when pain was rated by the Doloplus-2 scale, only residents' physical function influenced the agreement of pain report for both research assistants (RAs)-nurse and RAs-NA pairs. CONCLUSIONS Nurses' pain report across pain variables may be more reliable than NAs. Furthermore, factors influencing the agreement of pain assessed by self-report measurement or behavioural observations may be different, needing further investigation. CONCLUSIONS Pain inquiry and behavioural observation should be performed routinely as the fifth vital sign for developing effective care plans. Institutional caregivers should be used as alternative or complementary informants for assessing pain in older people with dementia.