The diagnosis of atlanto-occipital dislocation (AOD) remains problematic as a result of a lack of reliable radiodiagnostic criteria. In Part 1 of the AOD series, we showed that the normal occiput-C1 joint in children has an extremely narrow joint gap (condyle-C1 interval [CCI]) with great left-right symmetry. In Part 2, we used a CCI of 4 mm or greater measured on reformatted computed tomographic (CT) scans as the indicator for AOD and tested the diagnostic sensitivity and specificity of CCI against published criteria. The clinical manifestation, neuroimaging findings, management, and outcome of our series of patients with AOD are also reported. For diagnostic sensitivity, we applied the CCI criterion on 16 patients who fulfilled one or more accepted radiodiagnostic criteria of AOD and who showed clinical and imaging hallmarks of the syndrome. All 16 patients had plain cervical spine x-rays, head CT scans, axial cervical spine CT scans with reconstruction, and magnetic resonance imaging scans. The diagnostic yield and false-negative rate of CCI were compared with those of four published "standard" tests, namely Wholey's dens-basion interval, Powers' ratio, Harris' basion-axis interval, and Sun's interspinous ratio. The diagnostic value of "nonstandard" indicators such as cervicomedullary deficits, tectorial membrane and other ligamentous damage, perimedullary subarachnoid hemorrhage, and extra-axial blood at C1-C2 were also assessed. For diagnostic specificity, we applied CCI and the "standard" and "nonstandard" tests on 10 patients from five classes of non-AOD upper cervical injuries. The false-positive diagnostic rates for AOD of all respective tests were documented. The CCI criterion was positive in all 16 patients with AOD with a diagnostic sensitivity of 100%. Fourteen patients had bilateral AOD with disruption and widening of both OC1 joints. Two patients had unilateral AOD with only one joint wider than 4 mm. The abnormal CCI varied from 5 to 34 mm. Eight patients showed blatant left-right joint asymmetry in either CCI or anatomic conformation. The diagnostic sensitivities for the "standard" tests are as follows: Wholey's, 50%; Powers', 37.5%; Harris', 31%; and Sun's, 25%, with false-negative rates of 50, 62.5, 69, and 75%, respectively. The sensitivities for the "nonstandard" indicators are: tectorial membrane damage, 71%; perimedullary blood, 63%; and C1-C2 extra-axial blood, 75%, with false-negative rates of 29, 37, and 25%, respectively. Fifteen patients with AOD had occiput-cervical fusion. There were one early and two delayed deaths (19% mortality); two patients (12%) had complete or severe residual high quadriplegia, but 11 children (69%) enjoyed excellent neurological recovery. CCI was normal in all 10 patients with non-AOD upper cervical injuries with a diagnostic specificity of 100%. The false-positive rates for the four "standard" tests were: Sun's, 60%; Harris', 50%; Wholey's, 30%; and Powers', 10%; for the "nonstandard" indicator, the rates were: cervicomedullary deficits, 70%; tectorial membrane damage, 40%; C1-C2 extra-axial blood, 40%; and perimedullary blood, 30%. The CCI criterion has the highest diagnostic sensitivity and specificity for AOD among all other radiodiagnostic criteria and indicators. CCI is easily computed from reconstructed CT scans, has almost no logistical or technical distortions, can capture occiput-C1 joint dislocation in all three planes, and is unaffected by congenital anomalies or maturation changes of adjacent structures. Because CCI is the only test that directly measures the integrity of the actual joint injured in AOD and a widened CCI cannot be concealed by postinjury changes in the head and neck relationship, it surpasses others that use changeable landmarks.