To define and test the inter- and intra-rater reliability of a grading system for staging osteoarthritis (OA) of the ankle with magnetic resonance imaging (MRI) (Norwich Osteoarthritis of the Ankle MRI Score, NOAMS).The MRI features to be included in the score were defined by a multidisciplinary expert panel through a Delphi process. An anonymised randomised dataset of 50 MRI studies was created from patients with concurrent plain radiographs to include 10 ankles of each of the Kellgren-Lawrence grades 0 to 4. Two experienced musculoskeletal radiologists and two trainees scored each ankle MRI twice independently and blinded to the plain radiographs.The inter-rater kappa coefficient of agreement for cartilage disease was 0.88 (95% confidence interval [CI]: 0.85, 0.91) for experienced raters and 0.71 (95% CI: 0.67, 0.76) for trainees. Inter-rater agreement for subchondral bone marrow oedema and cysts varied from 0.73 to 0.82 for experienced raters and from 0.63 to 0.75 for trainees with lowest 95% CI of 0.48 and 0.63. When bone marrow lesions were combined into a total joint score the level of agreement increased to between 0.88 and 0.97 with lowest 95% CI of 0.86. Combining cartilage zone scores did not increase the reliability coefficients.An expert panel considered that cartilage degradation and subchondral bone marrow lesions were the most important features for staging the severity of ankle OA on MRI. Experienced observers can grade the severity of ankle OA on MRI with a clinically useful high degree of reproducibility.