Autopsy rates have declined steadily throughout the 1970s and 1980s. This trend, if not reversed, could negatively affect information provided to families as well as the training of physicians. In the 1980s, low autopsy rates among neonatal deaths, 50-72%, were attributed to limited parental understanding and to physicians' attitudes. In the early 1990s, alternatives to the routine autopsy, such as limited endoscopic/needle autopsies, were widely popularised, and the physician's education about autopsy vigorously addressed. The effects of these efforts on autopsy rates in the latter part of the 1990s have not been well studied. The study population consisted of 643 infants who died at an Ohio, USA, neonatal intensive care unit (NICU) between 1 January 1986 and 31 December 1998. Neonatologists obtained consent for autopsy. Information obtained from the computer database included birth hospital, gestational age, age at death, birth and death weight, race, sex, death year and maternal age, religion, gravidity and parity. The overall autopsy rate during the 13-year study period was 54% [95% CI 44.1%, 63.1%]. There was a trend towards progressive decrease in autopsy rates, 59% in 1986-89, 52% in 1990-94 and 47% in 1995-98 (P = 0.078). Autopsy was associated with increasing gestational age (P < 0.001) and decreasing parity [OR = 0.53 for >2 compared with <or=2, 95% CI 0.36, 0.99], but not with other variables. Autopsy rates remain suboptimum and may be declining despite efforts at educating the medical and general community about its benefits. Among infants who die in the NICU, autopsy is more likely the more mature the infant at birth and the fewer children the mother has. Physicians need to seek autopsy more conscientiously, with consideration for the emotional well-being of the family, and for its potential diagnostic and research value to science.