We assessed the diagnostic performance of magnetic resonance (MR) arthrography in the diagnosis of articular-sided partial-thickness and full-thickness rotator cuff tears in a large symptomatic population. MR arthrograms obtained in 275 patients including a study group of 139 patients with rotator cuff tears proved by arthroscopy and a control group of 136 patients with arthroscopically intact rotator cuff tendons were reviewed in random order. MR imaging was performed on a 1.0 T system (Magnetom Expert, Siemens). MR arthrograms were analyzed by two radiologists in consensus for articular-sided partial-thickness and full-thickness tears of the supraspinatus, infraspinatus, and subscapularis tendons. At arthroscopy, 197 rotator cuff tears were diagnosed, including 105 partial-thickness (93 supraspinatus, nine infraspinatus, three subscapularis) and 92 full-thickness (43 supraspinatus, 20 infraspinatus, 29 subscapularis) tendon tears. For full-thickness tears, sensitivity, specificity, and accuracy were 96%, 99%, and 98%, respectively, and for partial tears 80%, 97%, and 95%, respectively. False negative and positive assessments in the diagnosis of articular-sided partial-thickness tears were predominantly [78% (35/45)] observed with small articular-sided (Ellman grade1) tendon tears. MR arthrography is highly accurate in the diagnosis of full-thickness rotator cuff tears and is accurate in the diagnosis of articular-sided partial-thickness tears. Limitations in the diagnosis of partial-thickness tears are mainly restricted to small articular-sided tears (Ellman grade 1) due to difficulties in differentiation between fiber tearing, tendinitis, synovitic changes, and superficial fraying at tendon margins.