RATIONALE AND OBJECTIVES: In this retrospective case series, we utilize arthroscopy as the gold standard to determine if magnetic resonance imaging (MRI) of the knee can predict osteochondritis dissecans (OCD) lesion stability, the most important information to guide patient treatment decisions. It is hypothesized that the classification system of the International Cartilage Repair Society (ICRS) will allow for improved assessment of lesion grade and stability in OCD. MATERIALS AND METHODS: Routine MRI studies of 46 consecutive patients with arthroscopically proven OCD lesions (mean age: 23.7 years; 26 male, 16 female) were assessed by three radiologists who were blinded to arthroscopic results. Arthroscopic images were evaluated by two orthopedic surgeons in consensus. The OCD criteria of the ICRS were applied to arthroscopy and imaging interpretations. Inter-rater correlation statistics and accuracy of magnetic resonance (MR) grading with respect to arthroscopy were determined. RESULTS: Only 56% of the available MR reports assigned a label of stable or unstable to the lesion description. Of these, 58% of the lesions were deemed unstable and 42% were stable. Accuracy was 53% when reports addressed stability. Utilizing the ICRS classification system, for all three readers combined, the respective sensitivity, specificity, and accuracy of MR imaging to determine lesion stability were 70%, 81%, and 76%. When compared to the original MRI report, the overall accuracy increased from 53% to 76% when readers were given the specific criteria of the OCD ICRS classification. However, inter-reader variability remained high, with Krippendorf's alpha ranging from 0.48 to 0.57. CONCLUSIONS: In this paper, we utilize arthroscopy as the gold standard to determine if MRI can predict OCD lesion stability, the most important information to guide patient treatment decisions. To our surprise, the analysis of the existing radiology reports that addressed stability revealed an overall accuracy in defining OCD lesion stability of about 53%. The classification system of the ICRS, created by an international multidisciplinary, multi-expert consortium, did markedly improve the accuracy, but consistency among different readers was lacking. This retrospective study on OCD reporting and classification highlights the inadequacy of existing classification schemes, and emphasizes the critical need for improved diagnostic MRI protocols in musculoskeletal radiology in order to propel it toward evidence-based medicine.