Proxy and subject responses to survey questions about chronic conditions, health symptoms, and physical and instrumental functioning were compared to determine the extent of disagreement, the direction of nonrandom discrepancies (i.e., bias), and how disagreement and bias vary by proxy and subject characteristics. Subjects included 538 community-dwelling women aged 65 years and older who participated in the third home interview of a health survey in Baltimore, Maryland, 1986, and a self-designated proxy for each. The authors observed kappa values of > 0.6 (i.e., substantial to almost perfect agreement) for five of nine chronic conditions, no health symptoms, six of seven physical tasks of daily living, and seven of seven instrumental tasks of daily living. With few exceptions, proxies were more likely than subjects to report the presence of a condition, symptom, or functional problem. Variations in agreement and bias were noted by subject and proxy characteristics, with different patterns observed for different measurement areas. When using proxy reports in place of self-reports, it is important to evaluate the impact that using proxies has on study results.