Objective: To compare elder patients' and their informants' ratings of the elder's physical and mental function measured by a standard instrument, the Medical Outcomes Study Short Form 12 (SF-12). Methods: This was a randomized, cross-sectional study conducted at a university-affiliated community teaching hospital emergency department (census 65,000/year). Patients >69 years old, arriving on weekdays between 10 AM and 7 PM, able to engage in English conversation, and consenting to participate were eligible. Patients too ill to participate were excluded. Informants were people who accompanied and knew the patient. Elder patients were randomized 1:1 to receive an interview or questionnaire version of the SF-12. The questionnaire was read to people unable to read. Two trained medical students administered the instrument. The SF-12 algorithm was used to calculate physical (PCS) and mental (MCS) component scores. Oral and written versions were compared using analysis of variance. The PCS and MCS scores between patient-informant pairs were compared with a matched t-test. Alpha was 0.05. Results: One hundred six patients and 55 informants were enrolled. The patients' average (±SD) age was 77 ± 5 years; 59 (56%; 95% CI = 46% to 65%) were women. There was no significant difference for mode of administration in PCS (p = 0.53) or MCS (p = 0.14) scores. Patients rated themselves higher on physical function than did their proxies. There was a 4.1 (95% CI = 99 to 7.2) point difference between patients' and their proxies' physical component scores (p = 0.01). Scores on the mental component were quite similar. The mean difference between patients and proxies was 0.49 (95% CI = 3.17 to 4.16). The half point higher rating by patients was not statistically significant (p = 0.79). Conclusions: Elders' self-ratings of physical function were higher than those of proxies who knew them. There was no difference in mental function ratings between patients and their proxies. Switching from informants' to patients' reports in evaluating elders' physical function in longitudinal studies may introduce error.