The number of people who cannot be identified at the time of death, sometimes referred to as John or Jane Does, is unknown, and little is known about them as a group. The study's objectives were to estimate the number of annual unidentified deaths, to identify demographic characteristics associated with dying unidentified, to determine whether the rates of such deaths vary geographically or over time, and to better characterize the causes of death. This was a population-based surveillance study of data collected from death certificates from 1979 to 2004 in the U.S. Subjects were selected by the absence of name, date of birth, and Social Security Number on their certificates. Main outcome measures were distributions by age, sex, and underlying cause of death and rates by sex, race, year, and state of death. An average of 413 unidentified persons died each year. The peak year was 1987 with 691 deaths, a rate of 28.5 per 10 million people. The rate declined to 9.7 per 10 million in 2004. Most unidentified decedents were male (80.6%). Unidentified death rates were highest among black people and in the Southwest. Among deaths for which the cause was known, 82.7% were due to injuries. Among injury deaths, 31.8% were homicides. Improvement in identification technology may have reduced rates of unidentified death since the 1980s. In addition, variations in rates of unidentified decedents may reflect changes in risk factors such as homelessness and substance abuse.