A statistical model was developed based on Poisson regression of incidence of childhood leukaemia and non-Hodgkin's lymphoma (NHL) in relation to population mixing among all 119 539 children born 1969-1989 to mothers living in Cumbria, north-west England, (excluding Seascale). This model was used to predict the number of cases in Seascale (the village adjacent to the Sellafield nuclear installation) children, born 1950-1989 and diagnosed before 1993. After allowing for age, the incidence of acute lymphoblastic leukaemia (ALL) and NHL was significantly higher among children born in areas with the highest levels of population mixing, relative risk (RR) = 11.7 (95% confidence interval (CI) 3.2-43) and was highest among children of incomers. The model predicted up to 3.0 (95% CI 1.3-6.0) cases of ALL/NHL in children born in Seascale compared to six observed and 2.0 (95% CI 1.0-3.4) cases in children resident, but not born, in Seascale compared to two observed. Population mixing is a significant risk factor for ALL/NHL, especially in young children, accounting for over 50% of cases in Cumbria and most cases in Seascale.