To determine whether antidepressants are clinically effective and acceptable for the treatment of depression in people who also have a physical illness. The method used was a systematic review of all randomised controlled trials (found by computer and hand searches) comparing any antidepressant drug with placebo or no treatment, in depressed adults with a specified physical disorder. The main outcome measures are numbers of individuals who recover/improve at the end of the trial and, as a proxy for treatment acceptability, numbers who complete treatment. 18 studies were included, covering 838 patients with a range of physical diseases. 6 studies used SSRIs, 3 atypical antidepressants, and the remainder tricyclics. Patients treated with antidepressants were significantly more likely to improve than those given placebo: about 4 patients would need to be treated with antidepressants to produce one recovery from depression which would not have occurred had they been given placebo (NNT 4.2, 95% CI 3.2-6.4). Most antidepressants (tricyclics and SSRIs together, 15 trials) produced a small but significant increase in dropout (OR 1.66, 95% CI 1.14-2.40. NNH 9.8, 95% CI 5.4-42.9). The "atypical" antidepressant mianserin produced significantly less dropout than placebo. Trends towards tricyclics being more effective than SSRIs, but also more likely to produce dropout were noted. The review provides evidence that antidepressants, significantly more frequently than either placebo or no treatment, cause improvement in depression in patients with a wide range of physical diseases.