Limited evidence exists regarding the validity of clinical examination for the detection of shoulder pathology. We therefore wished to establish the sensitivity, specificity, positive predictive value and negative predictive value of clinical tests and magnetic resonance imaging (MRI) in the diagnosis of rotator cuff disorders against findings at arthroscopy. Using recognised tests for specific shoulder lesions, 117 patients with shoulder symptoms awaiting surgery were examined in a standard manner. The diagnoses were categorised and compared with abnormalities found on MRI and at surgery. Results were cross-tabulated to determine the above parameters. Ninety-four patients formed the study group with a mean age of 51 years. The median duration of symptoms was 45 weeks. For clinical examination, sensitivity and specificity to detect a tear or rupture of supraspinatus were 30 % (16/54) and 38 % (15/40) and, for the detection of any pathology, were 94 % (67/71) and 22 % (5/23), respectively, compared with arthroscopy. Correspondingly, the sensitivity of MRI compared with arthroscopy to detect a tear or rupture of supraspinatus was 90 % (28/31) with a specificity of 70 % (46/53), whereas for the detection of any abnormality, the sensitivity was 92 % (65/71) with a specificity of 48 % (11/23). The sensitivity of detecting any rotator cuff abnormality is high when examination and MRI is compared with arthroscopy with the specificity being greater with MRI than examination. In patients with shoulder symptoms severe enough to consider surgery, clinical assessment followed by specific imaging may help define the pathology in order to direct appropriate management.