Although the international and regional variability of perinatal mortality rates have been widely studied, less is known about the variability at the small-area level. The geographical distribution of perinatal mortality in the former North-West Thames Health Region, England, during 1981-90, and its association with small-area socio-economic factors, as measured by the Carstairs index of deprivation, were studied. Recently developed methods of analysis, including use of Bayesian statistics, were applied to obtain descriptive results and maps, and for fitting regression models that allowed for the presence of unmeasured risk factors. Significant heterogeneity (P < 0.001) of perinatal mortality across census wards and districts was found. The 5% of wards with the highest mortality experienced a risk more than 1.7 times that of the 5% with lowest mortality. Significant, positive association between deprivation and perinatal mortality was also found. Assuming causality, social differentials at the small-area level accounted for between 1.3% and 14.1% of all perinatal deaths, depending on which level of the Carstairs index was selected as reference. Although a proportion of such variability might be explained by social characteristics, a better understanding of the nature of the association is necessary.