We analyzed US fetal death and linked infant birth-death certificate data for 1995-1998 to evaluate perinatal deaths (late fetal deaths [> or = 28 weeks' gestation] and early neonatal deaths [< or = 7 days of life]) by race, Hispanic ethnicity, state of residence, and selected demographic characteristics. We also compared components of perinatal mortality, late fetal deaths, and early neonatal deaths, by birthweight, gestational age, and selected maternal medical conditions during pregnancy. From 1995 through 1998, there were 221,767 fetal deaths at > or = 20 weeks' gestation and infant deaths at less than 1 year. Of these, 113,421 (51%) were perinatal deaths; late fetal deaths accounted for 47% of perinatal deaths. The total perinatal mortality rate declined 5.3%, from 7.5 to 7.1 per 1,000 live births plus late fetal deaths. Blacks experienced higher perinatal mortality rates than whites (rate ratio = 2.1). Among perinatal deaths > or = 28 weeks' gestation, the ratio of fetal to neonatal deaths was 3.4 among blacks and 2.4 among whites. State-specific rates ranged from 5.2 to 13.1 per 1,000 live births plus late fetal deaths. Although late fetal deaths are not included in routine statistics of pregnancy outcomes, these deaths represent a large proportion of adverse pregnancy outcomes. Surveillance of perinatal mortality provides a more complete picture of the health of women, fetuses, and newborns. Improving the quality of surveillance data regarding fetal deaths is essential for more effective use of these data. This information can be used to prevent excess perinatal deaths and reduce disparities in pregnancy outcomes among high-risk subgroups identified by individual and population characteristics.