BACKGROUND: In this systematic review of randomized controlled trials (RCTs), we assess the benefits and harm of current treatments for diffuse proliferative lupus nephritis (DPLN). METHODS: The Cochrane Controlled Trial Registry, MEDLINE, and EMBASE were searched for RCTs of treatment for DPLN. All available RCTs of patients with biopsy-proven DPLN were included, and data were extracted for overall mortality, end-stage renal disease, doubling of serum creatinine level, relapse, major infection, herpes zoster infection, ovarian failure, malignancy, and bladder toxicity. Treatment effects on these outcomes were summarized as relative risk (RR) with 95% confidence interval (CI) and pooled by using a random-effects model. RESULTS: Twenty-five of 920 articles identified were eligible RCTs and were included. The majority compared cyclophosphamide or azathioprine plus steroids versus steroids alone. Cyclophosphamide plus steroids reduced the risk for doubling of serum creatinine level (4 RCTs, 228 patients; RR, 0.59; 95% CI, 0.40 to 0.88) compared with steroids alone, but had no impact on overall mortality (5 RCTs, 226 patients; RR, 0.98; 95% CI, 0.53 to 1.82). However, risk for ovarian failure was increased significantly (3 RCTs, 147 patients; RR, 2.18; 95% CI, 1.10 to 4.34). In studies from the 1970s, azathioprine plus steroids reduced the risk for all-cause mortality compared with steroids alone (3 RCTs, 78 patients; RR, 0.60; 95% CI, 0.36 to 0.99), but had no effect on renal outcomes. Neither therapy was associated with increased risk for major infection. The addition of plasma exchange to these treatments offered no benefit, and information on other agents, including mycophenolate mofetil, was insufficient for analysis. CONCLUSION: Until future RCTs of newer agents are completed, the current use of cyclophosphamide combined with steroids remains the best option to preserve renal function in patients with DPLN. The smallest effective dose and shortest duration of treatment should be used to minimize gonadal toxicity without compromising efficacy.