Abstract Objectives To determine the level of agreement between both proxy versions and the self-completed EQ-5D-5L. Design A randomized agreement study. Setting and participants We recruited 120 patients (compos mentis) and their proxies at the orthopaedic outpatient clinic. Patients completed the regular EQ-5D-5L and their proxy completed the proxy version of the EQ-5D-5L and rated the patients’ health from their own (proxy-proxy) perspective (i.e. how do you rate the health of the patient), and from the patient’s (proxy-patient) perspective (i.e. how do you think the patient would rate their own health if they were able to). Measures The primary outcome was the agreement between patients and their proxy, quantified as the intra class correlation coefficient for the EQ-5D-5L Utility score. Results Average Utility scores were 0.65 with the self completed EQ-5D-5L, versus 0.60 with the proxy-patient version and 0.58 with the proxy-proxy version. The ICC was 0.66 (95% CI 0.523, 0.753) for the proxy-patient perspective and 0.58 (95% CI 0.411, 0.697) for the proxy-proxy perspective. The mean gold standard score of the VAS-Health was 69.7 whereas the proxy-proxy perspective was 66.5 and the proxy-patient perspective was 66.3. Conclusion and implications The proxy-patient perspective yielded substantial agreement with the self completed EQ-5D-5L, while the agreement with the proxy-proxy perspective was moderate. In this study population of patients without cognitive impairment, proxies tended to underestimate the quality of life of their relative.