The purpose of this study was to measure interobserver and intraobserver variation in reporting magnetic resonance images of the temporomandibular joint. This study was based on magnetic resonance images of 149 joints. The images were interpreted independently by two of us on two occasions, 2 to 4 weeks apart. We were trained together before starting the study. The magnetic resonance images were interpreted for the position and configuration of the disk and osseous changes. The intraobserver agreement for reporting the position of the disk was 95%, configuration of the disk was 90%, and osseous changes was 97% (average of both observers). The agreement between the two observers was 90% on the position of the disk, 89% on the configuration of the disk, and 94% on reporting osseous changes (average of both readings). The results suggest that interobserver and intraobserver variation in reporting magnetic resonance images of the temporomandibular joint can be kept at a low level with high quality images, suitable training of the observers, and well-defined criteria for interpretation. The low observer variation together with the high diagnostic accuracy reported in other studies suggests that magnetic resonance imaging of the temporomandibular joint is a reliable and valid imaging technique for assessment of this joint.