BACKGROUND: Proximal humeral fractures are common yet the management of these injuries varies widely. In particular, the role and timing of any surgical intervention have not been clearly defined. OBJECTIVES: To collate and evaluate the scientific evidence supporting the various methods used for treating proximal humeral fractures. SEARCH STRATEGY: We searched the Cochrane Musculoskeletal Injuries Group specialised register, the Cochrane Central Register of Controlled Trials, PEDro, MEDLINE (1966 to May week 4 2003), EMBASE (1980 to 2003 week 22), CINAHL (1982 to May week 3 2003), AMED (1985 to May 2003), the National Research Register (UK), Current Controlled Trials, and bibliographies of trial reports. The search was completed in May 2003. SELECTION CRITERIA: All randomised studies pertinent to the treatment of proximal humeral fractures were selected. DATA COLLECTION AND ANALYSIS: Independent quality assessment and data extraction were performed by two reviewers. Although quantitative data from trials are presented, trial heterogeneity prevented pooling of results. MAIN RESULTS: Twelve randomised trials were included. All were small; the largest study involved only 86 patients. Bias in these trials could not be ruled out. Eight trials evaluated conservative treatment, three compared surgery with conservative treatment and one compared two surgical techniques. In the 'conservative' group there was very limited evidence indicating that the type of bandage used made any difference in terms of time to fracture union and the functional end result. However, an arm sling was generally more comfortable than a body bandage. There was some evidence that 'immediate' physiotherapy, without routine immobilisation, compared with that delayed until after three weeks immobilisation resulted in less pain and both faster and potentially better recovery in patients with undisplaced two-part fractures. Similarly, there was evidence that mobilisation at one week instead of three weeks alleviated pain in the short term without compromising long term outcome. Two trials provided some evidence that patients, when given sufficient instruction to pursue an adequate physiotherapy programme, could generally achieve a satisfactory outcome if allowed to exercise without supervision. Operative reduction improved fracture alignment in two trials. However, in one trial, surgery was associated with a greater risk of complication, and did not result in improved shoulder function. In one trial, hemi-arthroplasty resulted in better short-term function with less pain and less need for help with activities of daily living when compared with conservative treatment for severe injuries. Fracture fixation of severe injuries was associated with a high rate of re-operation in one trial, comparing tension-band wiring fixation with hemi-arthroplasty. There was very limited evidence that similar outcomes resulted from mobilisation at one week instead of three weeks after surgical fixation. REVIEWER'S CONCLUSIONS: Only tentative conclusions can be drawn from the available randomised trials, which do not provide sufficient evidence for many of the decisions that need to be made in contemporary fracture management. Early physiotherapy, without immobilisation, may be sufficient for some types of undisplaced fractures. It is unclear whether operative intervention, even for specific fracture types, will produce consistently better long term outcomes. There is a need for good quality evidence for the management of these fractures.