Observer variability in local radiological reading is a major concern in large-scale multi-center traumatic brain injury (TBI) studies. A central review process has been advocated to minimize this variability. The aim of this study is to compare central with local reading of TBI imaging datasets and to investigate the added value of central review. A total of 2050 admission computed tomography (CT) scans from subjects enrolled in the Collaborative European NeuroTrauma Effectiveness Research in Traumatic Brain Injury (CENTER-TBI) study were analyzed for seven main CT characteristics. Kappa statistics were used to calculate agreement between central and local evaluations and a center-specific analysis was performed. The McNemar test was used to detect whether discordances were significant. Central interobserver and intra-observer agreement was calculated in a subset of patients. Good agreement was found between central and local assessment for the presence or absence of structural pathology (CT+, CT-, κ = 0.73) and most CT characteristics (κ = 0.62 to 0.71), except for traumatic axonal injury lesions (κ = 0.37). Despite good kappa values, discordances were significant in four of seven CT characteristics (i.e., midline shift, contusion, traumatic subarachnoid hemorrhage, and cisternal compression; p = 0.0005). Central reviewers showed substantial to excellent interobserver and intra-observer agreement (κ = 0.73 to κ = 0.96), contrasted by considerable variability in local radiological reading. Compared with local evaluation, a central review process offers a more consistent radiological reading of acute CT characteristics in TBI. It generates reliable, reproducible data and should be recommended for use in multi-center TBI studies.