Abstract Background Community-acquired pneumonia (CAP) is one of the top causes of life-years lost globally. The optimal empiric antibiotic therapy regimen is uncertain. Randomized controlled trials (RCTs) provide useful information about relative antibiotic effectiveness. Methods We systematically searched Medline, EMBASE, and CENTRAL for RCTs comparing at least two empiric antibiotic regimens in patients with CAP, to March 17, 2017. We performed a systematic review and network meta-analysis and network meta-regression using a Bayesian framework. We used GRADE to assess certainty in the effect estimates. Results From 18,056 citations, we included 303 RCTs. Most studies (69.9%) were not blinded. All networks had low global heterogeneity (I2 0%). There were 26,423 participants included in the analysis of mortality and 30,559 for treatment failure. Seven hundred and twenty-six (2.9%) participants died. Patients randomized to third generation cephalosporins alone had higher mortality than those randomized to early generation fluoroquinolones (risk ratio [RR] 2.08, 95% credible interval 1.17–3.90), later generation fluoroquinolones (RR 2.32, 1.44–4.26), and cephalosporin-fluoroquinolone combinations (RR 3.21, 0.99–12.49). Participants who were randomized to a cephalosporin plus macrolide were less likely to die than those who received a third generation cephalosporin alone (RR 0.47, 0.21–0.99). The evidence was similar for treatment failure. Β-lactam plus β-lactamase inhibitors (e.g., piperacillin–tazobactam), early generation cephalosporins, and daptomycin appeared to confer a higher risk of mortality and/or treatment failure than most other antibiotic regimens including third-generation cephalosporins alone. For key comparisons, the GRADE quality of evidence was low or moderate. Conclusion In patients with CAP, an antibiotic regimen that includes a fluoroquinolone (and possibly a macrolide) may reduce mortality by ~1–2% compared with β-lactams (with or without a β-lactamase inhibitor) and cephalosporins alone. High quality, blinded and pragmatic randomized evidence would be helpful to increase certainty in the evidence. Disclosures All authors: No reported disclosures.