Cancer risk in children born before term has been assessed in a large number of case-control studies but very rarely in cohort studies. We carried out a cohort study of 35 178 children with the diagnosis immaturity at birth in the Hospital Discharge Register during 1977-89. The children were followed for cancer in the Danish Cancer Registry until 1994 and comparisons were made with incidence rates for all children in Denmark. The 64 observed cases of childhood cancer in the cohort corresponded closely to the expected number {standardised incidence ratio (SIR) = 1.03; [95% confidence interval (CI) 0.80, 1.32]}. The only cancer site with an observed number that deviated significantly from the expected number was central nervous system (CNS) tumours (26 cases observed; SIR = 1.57; [95% CI 1.02, 2.30]) in particular medulloblastoma (9 cases observed; SIR = 3.1; [95% CI 1.4, 5.9]). In a nested case-control study of the CNS tumours, we found that more cases than controls had been exposed to diagnostic X-rays, but the result was not significant. Surprisingly, for those born before term, the risk of CNS tumours increased with increasing gestational age in the nested case-control data. Our results are in line with previous evidence that children born before term are not at increased risk for childhood cancer in general. An explanation behind the excess of CNS tumours could not be identified, but the effect of diagnostic X-rays in newborns may deserve further attention.