BACKGROUND: Improving the health of expectant mothers and reductions in health inequalities, are repeatedly prioritised in policy reports in England and Northern Ireland. Measurement of underlying rates, and geographical variation in rates, of adverse birth outcomes are tools in monitoring these priorities. METHODS: Northern Ireland data on stillbirths, infant mortality and low birth weight (1992-2002) were linked to board (n=4), district council (n=26) and 1991 census wards (n=568). Underlying variations in rates were estimated at each geographical level, unadjusted and controlling for year, ward-level deprivation, settlement size and higher geographical levels. Impacts on geographical variation of individual social class, maternal age, multiple birth and smoking were assessed. RESULTS: There was significant variation in underlying rates of low birth weight (<2500 g) at all three geographical levels. Controlling for smoking reduced variation between wards. Geographical variation proved more robust for medium than for very low birth weight. No variation was seen between boards for other outcomes, nor between district level rates of infant mortality. Evidence was weak for variation in district rates of neonatal deaths and stillbirths, and variation in ward-level adjusted stillbirth rates was not significant. Variation in ward-level infant death rates was robust to all adjustments, with risks tripling (infant mortality) or quadrupling (neonatal mortality) between the 10th and 90th percentile. CONCLUSIONS: Strong evidence was found of geographical variation in infant mortality and low birth weight, unexplained by individual risk factors or by area-level deprivation. Geographical targeting or area-level interventions might look beyond deprivation scores, to other environmental and social factors.