The EUROCLUS study assembled incidence data for 13,551 cases of childhood leukaemia (CL) diagnosed between 1980 and 1989 in 17 countries (or regions of countries). These were referenced by location at diagnosis to small census areas of which there were 25,723 in the study area. Population counts, surface area and, hence, population density were available for all these small areas. Previous analyses have shown limited extra-Poisson variation (EPV) of case counts within small areas; this is most pronounced in areas of intermediate population density (150-499 persons/km2). In this study, the data set was examined in more detail for evidence that variations in incidence and EPV of CL are associated with population density. Incidence showed a curvilinear association with population density and was highest in areas which were somewhat more densely populated (500-750 persons/km2), where the incidence rate ratio relative to areas having > or = 1000 persons/km2 was 1.16 (95% confidence interval 1.07-1.26) and the P value for quadratic trend across eight strata of population density was 0.02. Incidence in these areas is uniformly elevated and showed no evidence of heterogeneity (i.e. EPV). Statistically significant evidence of EPV was evident amongst some of the areas previously classified as intermediate density areas (specifically, those with a density of 250-499 persons/km2, P < 0.001 for CL). These results were interpreted in terms of the current aetiological hypotheses for CL which propose that exposure to localised epidemics of one or more common infectious agent may contribute to the development of leukaemia. They suggest that such epidemics arise regularly in moderately densely populated areas and also sporadically in areas which are somewhat less densely populated. Although other interpretations are possible, these results may assist in the identification of characteristics which infectious agents must possess if direct or indirect causes of CL.