The objective of this study was to compare the image quality, sensitivity, specificity, and diagnostic accuracy of an open low-field MR system (0.2 T) with a standard high-field MR system (1.5 T) after arthrography of the shoulder. Thirty-eight patients either with suspected chronic instability (n = 12) or rotator cuff abnormalities (n = 26) were examined. Intra-articular injection of diluted Gd-DTPA was followed in randomized order either first by imaging on an open 0.2-T system or on a 1.5-T system. The image material was evaluated independently by two radiologists in a blinded fashion with respect to overall image quality and the detection of rotator cuff as well as capsular and labral abnormalities. Surgical correlation was available in 27 (71%) of 38 patients. For both systems, sensitivity and specificity for rotator cuff tears were 100% each, and for labrum pathologies, these values were 100 and 93%, respectively. The agreement for detection of labral pathologies between low-field and high-field examinations was good (kappa = 0.69, kappa = 0.61). For the detection of full-thickness tears of the rotator cuff, the agreement between the low-field and high-field MR examinations was very good and significant (kappa = 0.94, kappa = 1, p < 0.001). Overall image quality was rated good in 17 (45%) and fair in 21 (55%) of 38 cases on the 0.2-T MR system, and good in 32 (84%) and fair in 6 (16%) of 38 cases on the 1.5-T system. Motion artifacts were considered low in 24 (63%) and moderate in 14 (37%) of 38 cases for the 0.2-T system and low in 34 (89%) and moderate in 4 (11%) for 1.5-T system. Based on our results, low-field MR compares favorably to high-field MR in the detection of major abnormalities of the glenohumeral joint, at least when MR arthrography is used. Disadvantages are the duration of the examination and thus the risk of reduced image quality caused by motion artifacts.