Investigation of shoulder pain is important before surgical treatment. The presence or absence of a full-thickness rotator cuff tear (FTRCT) may determine the type of surgical treatment. Both MRI and conventional arthrography can be used, but little is known about their relative diagnostic and therapeutic impact. We performed a prospective trial assessing: (a) the influence of MRI and arthrography results on the clinician's diagnostic thinking (diagnostic impact); (b) the influence of the results on the clinician's therapeutic thinking (therapeutic impact); and (c) the diagnostic performance of the two techniques in patients with surgical confirmation. A total of 104 consecutive patients with shoulder problems referred to a specialist orthopaedic shoulder clinic underwent pre-operative investigation with MRI and arthrography. The surgeon's diagnosis, diagnostic confidence and planned treatment were measured before the investigation, and then again after the results of each investigation. Before the presentation of the investigation, results, the patients were randomised into two groups. In one group MRI was presented first; in the other group, arthrography. The MRI results led to fewer changes in diagnostic category (14 of 46, 30%) than arthrography (20 of 54, 37%), but the difference was not significant (P > 0.5). Magnetic resonance imaging led to slightly more changes in planned management (17 of 47, 36%) than arthrography (14 of 55, 25%), but again the difference was not statistically significant (P > 0.3). The results of the second investigation always had less diagnostic and therapeutic impact than the first. The accuracy of MRI for FTRCT in 38 patients with surgical confirmation was 79%, sensitivity 81% and specificity 78%; the accuracy of arthrography was 82%, sensitivity 50% and specificity 96%. The clinical diagnosis and management plan can be adequately defined by a single radiological investigation. Magnetic resonance imaging and arthrography had fairly similar diagnostic and therapeutic impact and comparable accuracy, although MRI was more sensitive and less specific. Magnetic resonance imaging may be the preferred investigation because of its better demonstration of soft tissue anatomy.