Environmental pollution as a cause of congenital anomalies is sometimes suspected because of clustering of anomalies in areas of higher exposure. This highlights questions around spatial heterogeneity (clustering) in congenital anomaly rates. If spatial variation is endemic, then any one specific cluster is less remarkable, though the presence of uncontrolled geographically clustered risk factors is suggested. If rates are relatively homogeneous across space other than around specific hazards, then evidence for these hazards causing the clusters is strengthened. We sought to estimate the extent of spatial heterogeneity in congenital anomaly rates in the United Kingdom.MethodsThe study population covered about one million births from five registers in Britain from 1991–1999. We estimated heterogeneity across four geographical levels: register area, hospital catchment, electoral ward, and enumeration district, using a negative binomial regression model. We also sought clusters using a circular scan statistic.ResultsCongenital anomaly rates clearly varied across register areas and hospital catchments (p <0.001), but not below this level (p > 0.2). Adjusting for socioeconomic deprivation and maternal age made little difference to the extent of geographical variation for most congenital anomaly subtypes. The two most significant circular clusters (of four ano-rectal atresias and six congenital heart diseases) contained two or more siblings.ConclusionThe variation in rates between registers and hospital catchment area may have resulted in part from differences in case ascertainment, and this should be taken into account in geographical epidemiological studies of environmental exposures. The absence of evidence for variation below this level should be interpreted cautiously in view of the low power of general heterogeneity tests. Nevertheless, the data suggest that strong localised clusters in congenital anomalies are uncommon, so clusters around specific putative environmental hazards are remarkable when observed. Negative binomial models applied at successive hierarchical levels provide an approach of intermediate complexity to characterising geographical heterogeneity.