Trends were examined in a cohort study of stillbirths and early and late neonatal deaths in Matlab, a rural area of Bangladesh between 1975 and 2002, using routinely collected demographic surveillance data. Main outcome measures were stillbirths per 1000 births, early neonatal deaths per 1000 livebirths, and late neonatal deaths per 1000 children surviving after 1 week. We performed a logistic regression examining trends over time and between two areas in the three outcome measures, controlling for the effects of parental education, religion, time, geography, parity, maternal age and birth spacing. There was a marked decline in stillbirths, early and late neonatal mortality over time in both areas, though the pace of decline was somewhat faster in the ICDDR,B (International Centre for Diarrhoeal Disease Research, Bangladesh) service area. Stillbirths declined by 24% overall in the ICDDR,B service area (crude OR comparing 1996-2002 with 1975-81: 0.76 [95% CI 0.68, 0.84]), compared with 15% in the Government service area (crude OR comparing 1996-2002 with 1975-81: 0.85 [0.76, 0.94]). The overall reduction in early and late neonatal mortality comparing the same periods was 39% and 73%, respectively, in the ICDDR,B area, compared with 30% and 63%, respectively, in the Government service area. Adjusting for socio-economic or demographic factors did not substantially alter the time or area differentials. The dramatic decline in neonatal mortality was, in large part, due to a fall in deaths from neonatal tetanus. The pace of decline was faster in the area receiving intense maternal and child health and family planning interventions, but stillbirths, early and late neonatal deaths also declined in the area not receiving such intense attention, suggesting that factors outside the formal health sector play an important role.