To determine the intraobserver and interobserver agreement on the geometric classification and 2-dimensional measurements of rotator cuff tears based on magnetic resonance arthrography. We retrospectively reviewed preoperative magnetic resonance arthrograms of 73 consecutive patients who were surgically treated for their full-thickness rotator cuff tears. The images were blinded and evaluated by 2 orthopaedic shoulder surgeons and 2 musculoskeletal radiologists using the geometric classification of rotator cuff tears (type 1, crescent-shaped tear; type 2, longitudinal U- or L-shaped tear; type 3a, massive tear measuring between 20 and 30 mm; and type 3b, massive contracted tear measuring >30 mm) and measuring the sagittal/coronal dimensions of the tear. Review was performed twice with an interval of at least 8 weeks. Agreement was calculated using the linear weighted κ coefficient and the intraclass correlation coefficient (ICC). The intraobserver agreement was excellent for both the geometric classification and the sagittal/coronal dimension measurement (κ, 0.81 to 0.92; ICC, 0.84 to 0.98). The ICC for the interobserver agreement was excellent for all sagittal and coronal dimension measurements (ICC, 0.95 to 0.97). The interobserver agreement for the geometric classification was good for the orthopaedic surgeons (κ, 0.75 for round 1 and 0.73 for round 2). The interobserver agreement for the radiologists was excellent in observation round 1 (κ, 0.82) and good in observation round 2 (κ, 0.71). The interobserver agreement between orthopaedic surgeons and radiologists was found to be moderate to good (κ, 0.52 to 0.66). The Fleiss κ was 0.66 for round 1 and 0.62 for round 2. The geometric classification and the 2-dimensional measurement of rotator cuff tears using magnetic resonance arthrography have good to excellent intraobserver agreement and moderate to good interobserver agreement among experienced observers. Level III, diagnostic study of nonconsecutive patients without consistently applied gold standard.