Objective: The objective of this study is to evaluate the the reliability of magnetic resonance imaging (MRI) in diagnosing alar ligament disruption in patients with potential atlanto-occipital dissociation (AOD). Materials and Methods: Three-blinded readers performed retrospective review on 6 patients with intra-operative confirmed atlanto-occipital dissocation in addition to a comparison cohort of patients with other cervical injuries that did not involve the atlanto-occipital articulation. Ligament integrity was graded from 1 to 3 as described by Krakenes et al. The right and left ligaments were assessed separately. Inter-observer agreement by patient, by group (AOD vs. non-AOD), and intra-observer agreement was calculated using weighted Cohen's kappa. Results: Interobserver agreement of alar ligament grade for individual patients ranged from slight to fair (κ = 0.05-0.30). Interobserver agreement of alar ligament grade for each group (AOD vs. non-AOD) ranged from fair to substantial (κ = 0.37-0.66). No statistically significant difference in categorical analysis of groups (AOD vs. non-AOD) and grade (0-1 vs. 2-3) was observed. Intraobserver agreement of individual patient's alar ligament grade ranged from moderate to substantial (κ = 0.50-0.62). Conclusion: The use of MRI to detect upper cervical ligament injuries in AOD is imperfect. Our results show inconsistent and unsatisfactory interobserver and intraobserver reliability in evaluation of alar ligament injuries. While MRI has immense potential for detection of ligamentous injury at the craniovertebral junction, standardized algorithms for its use and interpretation need to be developed.