Abstract Long-term sequelae are major limitations of radiation therapy use, especially for childhood brain tumour. Circle of Willis irradiation strongly increases the long-term risk of stroke, but to establish dose-response relationship, anticipating long-term effects of new techniques, requires to perform accurate and reproducible dosimetric estimations in large cohorts of patients having received radiotherapy decades ago. For the accuracy of retrospective dose reconstruction, the topographic variability of the Circle of Willis arteries is crucial. In order to improve retrospective dosimetric studies and dose-volume estimates to the typical Circle of Willis arteries, we aim to study the inter-individual topographic variability of these structures. Thirty-eight time of flight MRI sequences of children aged 2–17 years in both genders were investigated. A region growth algorithm was used for the segmentation of the cerebral arteries. A rigid registration in a common skull was performed following the anatomy of skull base foramina. The Posterior clinoid processes of the sella turcica were used as reference landmark (R0), and 5 key landmarks were chosen in each segmented Circle of Willis, then distances between the 5 landmarks and R0 were calculated for each of the 38 subjects. The distance between R0 and each landmark of the Circle of Willis followed a normal distribution, the average values ranging from 13.6 to 17.0 mm, and the standard deviations ranged from 2.6 to 3.0 mm, i.e. less than a fifth of the average value. The perimeter of the Circle of Willis was longer in older subjects, this increase being isotropic. Our study shows a remarkably low topographic variability of the typical Circle of Willis. An important result, allowing reliable anthropomorphic phantoms-based retrospective estimations of the radiation doses delivered to these arterial structures during radiotherapy treatment.