The New Zealand Cancer Registry (NZCR) is the main source of data on cancer incidence in New Zealand. The accuracy and completeness of registration of childhood cancers (ages zero to 14 years) were assessed during the conduct of a case-control study. Newly diagnosed children (1990-93) were ascertained from three sources: the NZCR; the Patient Management System (hospital admissions and discharges); and the separate Children's Cancer Registry. Pathology reviews were conducted to verify the diagnoses. Capture-recapture methods were used to assess the completeness of ascertainment. During the four-year period, 409 incident cases of childhood cancer were confirmed, giving an age-standardized incidence rate of 131 per million person-years (95 percent confidence interval = 119-144). The NZCR ascertained 395 (97 percent) of these children. In addition, the NZCR notified us of 43 other 'childhood cancer' registrations which were not confirmed as new cases of childhood cancer according to our eligibility criteria. The main reasons for these were coding errors (20 registrations), duplicates (seven), and a change in the pathological diagnosis as a result of the pathology review (seven). The capture-recapture estimate of the total number of incident cases was 410. Overall, the NZCR had good completeness for childhood cancers, but the number of unconfirmed registrations was larger than expected.