Objective To appraise the evidence for comparative efficacy and tolerability of drug treatments in patients with generalised anxiety disorder. Design Systematic review of randomised controlled trials. Primary Bayesian probabilistic mixed treatment metaanalyses allowed pharmacological treatments to be ranked for effectiveness for each outcome measure, given as percentage probability of being the most effective treatment. Secondary frequentist mixed treatment metaanalyses conducted with random effects model; effect size reported as odds ratio and 95% confidence interval. Data sources Medline, Embase, BIOSIS, PsycINFO, Health Economic Evaluations Database, National Health Service Economic Evaluation Database, and Database of Abstracts of Reviews of Effects via DataStar, and Cochrane Database of Systematic Reviews via Cochrane Library (January 1980 to February 2009). Eligibility criteria Double blind placebo controlled randomised controlled trials; published systematic reviews and meta-analyses of randomised controlled trials. Randomised controlled trials including adult participants (aged ≥18) receiving any pharmacological treatment for generalised anxiety disorder. Data abstraction methods Titles or abstracts reviewed initially, followed by review of full text publications for citations remaining after first pass. A three person team conducted screening; an independent reviewer checked a random selection (10%) of articles screened. Data extracted for meta-analysis were also independently reviewed. Main outcome measures Proportion of participants experiencing ≥50% reduction from baseline score on Hamilton anxiety scale (HAM-A) (response), proportion with final HAM-A score ≤7 (remission), proportion withdrawing from trial because of adverse events (tolerability). Results The review identified 3249 citations, and 46 randomised controlled trials met inclusion criteria; 27 trials contained sufficient or appropriate data for inclusion in the analysis. Analyses compared nine drugs (duloxetine, escitalopram, fluoxetine, lorazepam, paroxetine, pregabalin, sertraline, tiagabine, and venlafaxine). In the primary probabilistic mixed treatment meta-analyses, fluoxetine was ranked first for response and remission (probability of 62.9% and 60.6%, respectively) and sertraline was ranked first for tolerability (49.3%). In a subanalysis ranking treatments for generalised anxiety disorder currently licensed in the United Kingdom, duloxetine was ranked first for response (third across all treatments; 2.7%), escitalopram was ranked first for remission (second across all treatments; 26.7%), and pregabalin was ranked first for tolerability (second across all treatments; 7.7%). Conclusions Though the frequentist analysis was inconclusive because of a high level of uncertainty in effect sizes (based on the relatively small number of comparative trials), the probabilistic analysis, which did not rely on significant outcomes, showed that fluoxetine (in terms of response and remission) and sertraline (in terms of tolerability) seem t...