We undertook a prospective study of 61 children in Malawi with septic arthritis of the shoulder. They were randomised into two groups, treated by aspiration (group 1, 31 patients) or arthrotomy (group 2, 30 patients). Both received antibiotics for six weeks. We studied the results of blood tests, microbiology, and the clinical and radiological outcome one year after diagnosis. Only one patient was sickle-cell positive and three were HIV-positive. Non-typhoidal Salmonella species accounted for 86% (19/22) of the positive joint cultures in group 1 and 73% (16/22) in group 2. Of the 33 radiographs available for review at follow-up at six months, 23 (70%) showed evidence of glenohumeral damage. There was no statistical difference in radiological outcome for the two groups. We devised and validated a scoring system, the Blantyre Septic Joint Score, for the assessment of joints based upon swelling, tenderness, function and range of movement. Despite the radiological changes only one of the 24 joints examined at one year had any deficit in these parameters. There was no statistical difference in the clinical outcome for the two treatment groups at any stage during the period of follow-up.