The aims of the study were to use the population base of New South Wales (NSW) to study all births from 20 to 27 weeks' gestation in 1992-3 and to compare two data sources for perinatal deaths. The prospective population-based statewide audit (NICUS) of infants admitted to tertiary neonatal intensive care units (NICUs) in NSW was used to collect data on infants less than 28 weeks' gestation registered in 1992-3. This audit also surveyed the 160 obstetric hospitals in NSW to ascertain information on stillbirths and early neonatal deaths in the study period. The NSW midwives data collection (MDC) was the other source of information on stillbirths and labour ward deaths. Data were analysed using SAS. In 1992-3 in NSW 1170 infants were born at 20-27 weeks' gestation. There were 556 stillbirths and 614 live births, of whom 180 (29.3%) died in the labour ward and 434 (70.7%) were admitted to a tertiary NICU. Sixty-six per cent of stillbirths were identified by both data collections, 16.5% by the MDC only and 17% by NICUS only. There was a high major congenital anomaly rate (18.5%) among the stillbirths. Two-thirds of the infants admitted to NICUs survived to 1 year. Information was available on at least one follow-up parameter for 89% (255/288) of the survivors to 12 months (corrected age). Of the 244 infants who had a neurological assessment by a paediatrician, 17% were diagnosed to have cerebral palsy. Eleven per cent of the 239 who had a formal Griffiths developmental assessment had a major intellectual disability. Five (2% of 255) of the 1-year-olds were blind, and 12 (4.7% of 255) had bilateral hearing aids. Seventy-one per cent of the infants examined at 1 year did not have a major disability. For accurate perinatal death data, collection from more than one source is recommended. Infants born at 20-27 weeks' gestation contribute 40% of all stillbirths in NSW, most of the costs of neonatal intensive care as well as the costs of long-term morbidity. In Australia in the early 1990s, the survival of infants born at less than 28 weeks' gestation was best from 26 weeks gestational age onwards. Long-term morbidity did not change from that of earlier cohorts. The most common major disability was cerebral palsy.