The objective of this study was to review the effectiveness and safety of antidepressants in neuropathic pain. In a systematic review of randomised controlled trials, the main outcomes were global judgements, pain relief or fall in pain intensity which approximated to more than 50% pain relief, and information about minor and major adverse effects. Dichotomous data for effectiveness and adverse effects were analysed using odds ratio and number needed-to-treat (NNT) methods. Twenty-one placebo-controlled treatments in 17 randomised controlled trials were included, involving 10 antidepressants. In six of 13 diabetic neuropathy studies the odds ratios showed significant benefit compared with placebo. The combined odds ratio was 3.6 (95% CI 2.5-5.2), with a NNT for benefit of 3 (2.4-4). In two of three postherpetic neuralgia studies the odds ratios showed significant benefit, and the combined odds ratio was 6.8 (3.5-14.3), with a NNT of 2.3 (1.7-3.3). In two atypical facial pain studies the combined odds ratio for benefit was 4.1 (2.3-7.5), with a NNT of 2.8 (2-4.7). Only one of three central pain studies had analysable dichotomous data. The NNT point estimate was 1.7. Comparisons of tricyclic antidepressants did not show any significant difference between them; they were significantly more effective than benzodiazepines in the three comparisons available. Paroxetine and mianserin were less effective than imipramine. For 11 of the 21 placebo-controlled treatments there was dichotomous information on minor adverse effects; combining across pain syndromes the NNT for minor (noted in published report) adverse effects was 3.7 (2.9-5.2). Information on major (drug-related study withdrawal) adverse effects was available from 19 reports; combining across pain syndromes the NNT for major adverse effects was 22 (13.5-58). Antidepressants are effective in relieving neuropathic pain. Compared with placebo, of 100 patients with neuropathic pain who are given antidepressants, 30 will obtain more than 50% pain relief, 30 will have minor adverse reactions and four will have to stop treatment because of major adverse effects. With very similar results for anticonvulsants it is still unclear which drug class should be first choice. Treatment would be improved if we could harness the dramatic improvement seen on placebo in some of the trials.