Agreement between self reports and proxy reports of health-related quality of life (H RQoL) was examined in a sample of 76 patients with mild to moderate Alzheimer's disease and their proxies. Patients and proxies completed an '17-item Duke health profile'. The items were rephrased for the proxy. The proportion of exact agreement between patients and proxies on the 17 items ranged from 26.3 to 52.6%. Results reveal poor to moderate agreement (intraclass correlation coefficients (ICCs) from 0.00 to 0.61 for 10 subscales) between patients' and proxies' reports. Agreement was higher for measures of function that are directly observable (physical health, disability) and relatively poor for more subjective measures. Proxy reliability varied according to the relationship of the proxy to the index subject. Spouses and nurses agreed more closely with index subjects than did children or nurses' aides. Agreement decreased with increasing severity of dementia. Statistically significant differences in mean scores were noted for several dimensions, with proxies tending to rate the patients as having a lower quality of life than the patients themselves. This study indicates the importance of considering the information source of a patient's HRQoL. However, assessments by proxies should be used with caution.