Background: Computed tomography (CT) has been used increasingly worldwide over the last decades. However, concerns have been raised about potential radiation-related cancer risks, particularly after exposure to CT in childhood, due to the greater radiation sensitivity of children and to their longer life expectancy allowing to develop radiation associated late health effects. Several CT cohort studies aim at understanding the relationship between CT organ dose and cancer incidence. However, several sources of uncertainty coming from CT acquisition parameters and patient’s morphology exist but have sparsely been accounted for in risk estimates. This may lead to biased risk estimates and misleading conclusions. In this work, the aim is to study the impact of CT organ dose uncertainties on the risk of central nervous system (CNS) tumors and leukemia in the French CT cohort. Methods: The French CT cohort includes almost 100,000 children who received at least one CT between 2000 and 2011 in one of the 21 participating university hospitals. Patients were followed until the first diagnosis of cancer, the death, the 18th birthday or the 31st December 2016. Examinations and radiological protocols were retrieved to estimate cumulative absorbed doses to the brain and the red bone marrow (RBM) with NCICT 1.2. Sensitivity analysis indices were computed to identify the most influential input parameters in the estimation of organ doses. A first submodel was proposed to describe the discrepancy between the CT values provided by the French radiological protocols for these input parameters and their “true” values. A second submodel was proposed to describe the discrepancy between the NCICT estimated organ doses and “true” dose values accounting for patients’ morphology. Cox and excess hazard ratio survival models were considered as two alternative doseresponse models. The submodels were combined into a unique framework and fitted simultaneously using a Bayesian learning algorithm to estimate cancer risks. PACS data (Picture archiving and communication system) were used to learn about dosimetric uncertainties. Results and discussion: When not accounting for dosimetric uncertainties, a Cox model-based frequentist estimation, with asymptotic 95% confidence intervals,showed statistically significant doseresponse relationships for CNS tumors and leukemia. Bayesian risk estimates were similar but with wider credible intervals which seems to call into question the use of any statistical estimation approach based on asymptotic assumptions, in this context of weak signal data. Moreover, we will show how dose uncertainties impact cancer risk estimates and compare the fitting abilities of several models