Objective This study aimed to investigate the reliability and validity of a self‐administered version of DEMQOL‐Proxy, a disease‐specific instrument that measures health‐related quality of life in people with dementia. Methods The sample consisted of 173 informal carers of people with dementia, aged 29 to 89 years old. Carers were mostly female, White/White British and closely related to the patient. They completed DEMQOL‐Proxy (self‐administered), EQ‐5D‐3L (proxy reported about the person with dementia), EQ‐5D‐3L (self‐reported about their own health) and the Zarit Burden Interview. Using well‐established methods from classical test theory, we evaluated scale level acceptability, reliability and convergent, discriminant and known‐groups validity of DEMQOL‐Proxy. Results DEMQOL‐Proxy (self‐administered) showed high acceptability (3.5% missing data and 0% scores at floor or ceiling), high internal consistency reliability (α = 0.93) and good convergent and discriminant validity. Amongst others, we found a moderately high correlation with EQ‐5D‐3L proxy reported (r = 0.52) and low to essentially zero correlations with EQ‐5D‐3L self‐reported (r = 0.20) and carer and patient background variables (r ≤ 0.20). As predicted, DEMQOL‐Proxy (self‐administered) showed a modest correlation with DEMQOL (r = 0.32). Known‐groups differences on health‐related quality of life (comparing people with versus people without cognitive impairment) were of moderate effect size (d = 0.38) and in the expected direction. Conclusions DEMQOL‐Proxy (self‐administered) has comparable acceptability, reliability and validity with DEMQOL‐Proxy (interviewer administered). DEMQOL‐Proxy (self‐administered) can be used in a wider variety of contexts than its interviewer‐administered version, including routine use in busy clinics. Copyright © 2016 John Wiley & Sons, Ltd.