BACKGROUND Osteoarthritis (OA) is a common joint disorder. In the knee, injections of corticosteroids into the joint (intraarticular (IA)) may relieve inflammation, and reduce pain and disability. OBJECTIVE To evaluate the efficacy and safety of IA corticosteroids in treatment of OA of the knee. METHODS We searched the Cochrane Central Register of Controlled Trials (CENTRAL) (Issue 2, 2003), MEDLINE (to January (week 1) 2006 for update), EMBASE, PREMEDLINE (all to July 2003), and Current Contents (Sept 2000). Specialised journals, trial reference lists and review articles were handsearched. METHODS Randomised controlled trials of IA corticosteroids for patients with OA of the knee: single/double blind, placebo-based/comparative studies, reporting at least one core OMERACT III outcome measure. METHODS Methodological quality of trials was assessed, and data were extracted in duplicate. Fixed effect and random effects models, giving weighted mean differences (WMD), were used for continuous variables. Dichotomous outcomes were analysed by relative risk (RR). RESULTS Twenty-eight trials (1973 participants) comparing IA corticosteroid against placebo, against IA hyaluronan/hylan (HA products), against joint lavage, and against other IA corticosteroids, were included.IA corticosteroid was more effective than IA placebo for pain reduction (WMD -21.91; 95% confidence interval (CI) -29.93 to -13.89) and patient global assessment (the RR was 1.44 (95% CI 1.13 to 1.82)) at one week post injection with an NNT of 3 to 4 for both, based on n=185 for pain on 100 mm visual analogue scale (VAS) and n=158 for patient global assessment. Data on function were sparse at one week post injection and neither statistically significant nor clinically important differences were detected. There was evidence of pain reduction between two weeks (the RR was 1.81 (95% CI 1.09 to 3.00)) to three weeks (the RR was 3.11 (95% CI 1.61 to 6.01), but a lack of evidence for efficacy in functional improvement. At four to 24 weeks post injection, there was lack of evidence of effect on pain and function (small studies showed benefits which did not reach statistical or clinical importance, i.e. less than 20% risk difference). For patient global, there were three studies which consistently showed lack of effect longer than one week post injection. However, all were fairly small sample sizes (less than 50 patients per group). This was supported by another study which did not find statistically significant differences, at any time point, on a continuous measure of patient global assessment (100 mm VAS).In comparisons of corticosteroids and HA products, no statistically significant differences were in general detected at one to four weeks post injection. Between five and 13 weeks post injection, HA products were more effective than corticosteroids for one or more of the following variables: WOMAC OA Index, Lequesne Index, pain, range of motion (flexion), and number of responders. One study showed a difference in function between 14 to 26 weeks, but no differences in efficacy were detected at 45 to 52 weeks. In general, the onset of effect was similar with IA corticosteroids, but was less durable than with HA products. Comparisons of IA corticosteroids showed triamcinolone hexacetonide was superior to betamethasone for number of patients reporting pain reduction up to four weeks post injection (the RR was 2.00 (95% CI 1.10 to 3.63). Comparisons between IA corticosteroid and joint lavage showed no differences in any of the efficacy or safety outcome measures. CONCLUSIONS The short-term benefit of IA corticosteroids in treatment of knee OA is well established, and few side effects have been reported. Longer term benefits have not been confirmed based on the RevMan analysis. The response to HA products appears more durable. In this review, some discrepancies were observed between the RevMan 4.2 analysis and the original publication. These are likely the result of using secondary rather than primary data and the statistical methods available in RevMan 4.2. Future trials should have standardised outcome measures and assessment times, run longer, investigate different patient subgroups, and clinical predictors of response (those associated with inflammation and structural damage).