The aim of this retrospective observational study was to assess the potential agreement between independent magnetic resonance imaging (MRI) and arthroscopic findings and their respective contributions to a final diagnosis in patients with refractory temporomandibular joint disorders. Two dentomaxillofacial radiologists and two oral and maxillofacial surgeons scored 50 joints. All observers, who were blinded to additional clinical information, used a specific scoring form and selected one or more diagnostic labels. Agreement between MRI and arthroscopy and their contributions to the final diagnosis were assessed as primary outcomes using Fleiss' kappa. Intra-modality agreement and the correlation between signal intensity ratio (SIR) measurements on MRI and synovitis grading on arthroscopy were assessed as secondary outcomes. Agreement between MRI and arthroscopy was poor. A fair level of agreement was only reached for reduction capacity of the disc and disc perforation. Arthroscopic diagnostic labels matched better with the final diagnosis, suggesting a bigger contribution to that diagnosis. Higher SIR measurements correlated with higher synovitis grading scores for the retrodiscal tissue and the posterior band of the disc. Intra-modality agreement was better in arthroscopy. When blinded to clinical information, arthroscopy and MRI observations can lead to different conclusions. The diagnostic outcomes of both examinations should be considered and integrated into a final diagnosis.