Objective: To compare contrast leakage, pain score, image quality and diagnostic performance at different doses and levels of local anaesthesia for direct shoulder magnetic resonance arthrography. Methods: Patients (n = 157) were prospectively enrolled and allocated to Group 1 (no local anaesthetic), Group 2 (local anaesthesia to subcutaneous fat level; lidocaine 1-2 ml), Group 3 (to deltoid muscle level; 3-5 ml), or Group 4 (to subscapularis muscle level; 6-8 ml). We evaluated the frequency of contrast leakage, periprocedural/postprocedural pain, contrast-To-noise ratio of the intra-Articular signal, and subjective image noise/image sharpness. Radiological diagnoses of superior anterior-To-posterior (SLAP) and Bankart lesions were assessed. All data were analysed by one-way analysis of variance/Kruskal-Wall, Χ2/Fisher's exact and DeLong's tests. Results: The frequency of contrast leakage from the injection path and subjective image noise were significantly lower in Groups 1 and 2 than in Groups 3 and 4 (p = 0.001- 0.04). Periprocedural/postprocedural pain scores among Groups 2-4 were similar and lower than those of Group 1. The contrast-To-noise ratio (p = 0.11-0.97) and image sharpness (p = 0.12-0.43) were similar among Groups 2-4 and significantly lower than those of Group 1 (p = 0.001- 0.02). The diagnostic performance for the assessment of superior anterior-To-posterior and Bankart lesions was better in Groups 2-4 than in Group 1, although there were no significant differences (p = 0.23-0.99). Conclusion: Local anaesthesia with 1-2 ml lidocaine at subcutaneous fat level reduced pain and provided optimal image quality in direct shoulder magnetic resonance arthrography. Advances in knowledge: This method can increase image quality, reduce periprocedural/postprocedural pain and potentially reduce the need for re-examination.