This study was developed to examine whether the use of a proxy respondent influences the distribution and factors associated with health perception among older adults, as well as the mortality relative risk. The study included 28,943 and 35,040 participants from the Brazilian National Household Survey (PNAD) in 1998 and 2003, respectively, and 1,606 members of the cohort from Bambuí, Minas Gerais, Brazil. The results showed that use of proxy in the PNAD 1998 and 2003 did not modify the distribution of health perception or associated factors. The 5-year mortality rate ratio among cohort participants was higher in direct interviews with seniors (RR = 2.40) as compared to those answered by proxies (RR = 1.28), but the direction of the association was the same. These results indicate that health perception data from the PNAD can be used regardless of who answers the interview and that the association between worse health perception and mortality persisted even when perception was inferred from a proxy.