AIMS: To evaluate the comparative efficacy (50% reduction in seizure frequency) and tolerability (premature withdrawal due to adverse events) of anti-epileptic drugs (AEDs) for refractory epilepsy. METHODS: We searched Cochrane Central Register of Controlled Trials (Cochrane Library 2009, issue 2) including Epilepsy Group's specialized register, MEDLINE (1950 to March 2009), EMBASE (1980 to March 2009), and Current Contents Connect (1998 to March 2009) to conduct a systematic review of published studies, developed a treatment network and undertook a network meta-analysis. RESULTS: Forty-three eligible trials with 6346 patients and 12 interventions, including placebo, contributed to the analysis. Only three direct drug comparator trials were identified, the remaining 40 trials being placebo-controlled. Conventional random-effects meta-analysis indicated all drugs were superior in efficacy to placebo (overall odds ratio (OR] 3.78, 95% CI 3.14, 4.55) but did not permit firm distinction between drugs on the basis of the efficacy or tolerability. A Bayesian network meta-analysis prioritized oxcarbazepine, topiramate and pregabalin on the basis of short term efficacy. However, sodium valproate, levetiracetam, gabapentin and vigabatrin were prioritized on the basis of short-term efficacy and tolerability, with the caveat that vigabatrin is recognized as being associated with serious visual disturbance with chronic use. CONCLUSION: Of the wide range of AEDs licensed for the treatment of refractory epilepsy, sodium valproate, levetiracetam and gabapentin demonstrated the best balance of efficacy and tolerability. Until regulators mandate greater use of active comparator trials with longer term follow-up, network meta-analysis provides the only available means to quantify these clinically important parameters.