Objective To introduce the TOI classification system (the Traction reduction type [T type], Operation reduction type [O type], and Irreducible type [I type] classification system) and to determine the interobserver and intraobserver reliability of the TOI classification system. Methods Based on the characteristics of atlantoaxial dislocation (AAD), AAD was divided into Traction reduction type (T type), Operation reduction type (O type), and Irreducible type (I type). The analysis of interobserver and intraobserver agreements was done using kappa statistics. From July 2016 to January 2019, 213 AAD patients were retrospectively studied at four hospitals. Plain radiographs including extension and flexion views and three-dimensional CT images were obtained. Twenty independent observers, including eight experienced spine specialists and 12 orthopaedic fellows from four different residency training hospitals, completed the survey. Results The classification of the TOI system was based on etiology, the course of the disease, flexion-extension X-rays, three-dimensional CT reconstruction, and curative effects of skull traction. Flexion-extension X-rays demonstrating a successful reduction of the dislocated atlantoaxial joint and three-dimensional CT images showing osseous fusion of atlantoaxial facet joints and cervical traction reveal characteristics of T-type. Furthermore, this type can be divided into two subtypes, T1 and T2, according to the etiology and course of the disease. Unsatisfactorily reduction after 1-2 weeks of strict cervical traction, no reduction shown on flexion-extension X-rays, and no destruction or boneless fusion of atlantoaxial facet joints demonstrated in three-dimensional CT images are characteristics of type O. Atlantoaxial facet joint showing bone fusion or failure of reduction after cervical traction or three-dimensional CT images showing failure of surgical release are characteristics of type I. Interobserver and intraobserver reliability of the TOI classification system were moderate (kappa = 0.543) and substantial (kappa = 0.658), respectively. Interobserver and intraobserver reliability of the treatment choice were moderate (kappa = 0.568) and substantial (kappa = 0.675), respectively. There were no significant differences in the interobserver and intraobserver reliability between experienced spine specialists and fellows for all kappa-values (P > 0.05). Conclusions The TOI classification system had satisfactory reliability and, therefore, can be applied clinically and used by less experienced surgeons. We believe TOI can help surgeons choose appropriate treatment strategies.