We examined consistency in the classification of ancestry by self, proxy, interviewer, and funeral director (on a death certificate) in a sample of the U.S. population--the First National Health and Nutrition Examination Survey and Epidemiologic Follow-up. Among study subjects for whom comparable ethnic identity options were available at both interviews, 58% of subjects specified the same identity at two times. Persons who specified four different ethnic backgrounds were 3.4 times as likely to change their identity over time as persons specifying only one background. Self-classification of ancestry at initial interview was consistent with proxy reports at follow-up for 55% of subjects for whom proxy information was available. Comparison of the self-classification of ancestry with the classification of race by interviewers and by funeral directors indicates high consistency for Whites and Blacks and low consistency for American Indians. The "measurement" of ancestry (that is, race or ethnicity) is critical to the understanding and elimination of differences in health status among racial/ethnic populations, but the low reliability of these measures over time and across observers complicates the analysis and interpretation of health statistics by ancestry, particularly for populations other than White or Black.