OBJECTIVE: Proxy measures are an alternative source of data for care home residents who are unable to complete the health utility measure, but the agreement levels between residents and care home staff for the EQ-5D have not been investigated previously. The objective of the present study was to examine the inter-rater agreement levels for the reporting of EQ-5D by care home residents and staff, adjusting for the impact of clustering. STUDY DESIGN AND SETTING: The data consist of EQ-5D scores for 565 pairs of care home residents and proxies and quality-adjusted life-years (QALYs) for 248 pairs. Cluster-adjusted agreement was compared for the domains, index scores, and QALYs from the EQ-5D. Factors influencing index score agreement are also described. RESULTS: The results show poor to fair agreement at the domain level (cluster-adjusted Kappa -0.03 to 0.26) and moderate agreement at the score level (cluster-adjusted intra-class correlation coefficient [ICC] 0.44-0.50) and for QALYs (cluster-adjusted ICC 0.59). A higher likelihood of depression and lower cognitive impairment were both associated with smaller discrepancy between proxy and self-completed scores. CONCLUSION: Proxies appear to be an acceptable source of data for index scores and QALYs but may be less reliable if individual domains are considered.