The EUROCLUS project included information on residence at diagnosis for 13351 cases of childhood leukaemia diagnosed in the period 1980-89 in defined geographical regions in 17 countries. A formal algorithm permits identification of small census areas as containing case excesses. The present analysis examines spatial-temporal patterns of the cases (n = 970) within these clustered areas. The objectives were, first, to compare these results with those from an analysis conducted for UK data for the period 1966-83, and, second, to extend them to consider infant leukaemias. A modification of the Knox test investigates, within the small areas, temporal overlap between cases in a subgroup of interest at a putative critical time and all other cases at any time between birth and diagnosis. Critical times were specified in advance as follows: for cases of acute lymphoblastic leukaemia aged 2-4 years, the 18-month period preceding diagnosis; for cases of total leukaemia aged 5-14 years, 1 year before to 1 year after birth; and for infant cases (diagnosed < 1 year), 1 year before to 6 months after birth. Each of the analyses found evidence of excess space-time overlap compared with that expected; these were 10% (P = 0.005), 15% (P= 0.0002) and 26% (P= 0.03) respectively. The results are interpreted in terms of an infectious origin of childhood leukaemia.