Objective: To identify the etiology of hemiplegic shoulder pain by magnetic resonance (MR) arthrography. Method: The study included seventy-four hemiplegic patients with shoulder pain. After several physical examinations, all patients had fluoroscopically guided injection by a physiatrist with a maximum of 12∼15 ml of contrast agent. Then T1-weighted, T2-weighted and fat-suppressed T1-weighted images were taken at the oblique coronal plane. In addition, fat-suppressed T1-weighted images were obtained at the oblique sagittal and oblique coronal plane. Results: Except for the 9 patients who did not finish the study, the mean age of the participants was 61.5±8.9 years and mean duration of the cerebrovascular accident (CVA) was 15.7±9.7 weeks. The findings were as follows: 40% supraspinatus tendinitis, 30.8% superior labrum anterior to posterior (SLAP) lesion, 29.2% adhesive capsulitis, 24.6% supraspinatus partial tear, 23.1% biceps tendinitis, 13.8% supraspinatus full thickness tear, 7.7% infraspinatus partial tear. The SLAP lesion had significant statistic relationship with biceps tendinitis (p＜0.05) but not with rotator cuff lesion. Conclusion: We found that causes of hemiplegic shoulder pain were various. The prevalence of the SLAP lesion was high (30.8%). We recommend the MR arthrography when the hemiplegic shoulder pain does not improve by conventional therapy or the cause of the pain is uncertain. (J Korean Acad Rehab Med 2008; 32: 657-663)