Objective – To consider some of the problems that may arise when the results of clinical trials are applied uncritically to everyday clinical practice. To show how an epidemiological perspective makes it easier to identify these problems. Method – A review of clinical trials to illustrate problems of: (a) covert selection bias in individual trials, (b) bias in the selection of trials included in reviews and meta-analyses. Results – Several kinds of selection bias are known to exist. In particular, trials with negative results are less likely to be published in widely read journals than are trials of similar scientific quality reporting positive results. Conclusions – There are dangers in applying the results of trials to clinical practice in an uncritical way. These dangers can be avoided by adopting and epidemiological perspective, seeking for clinical trials that are representative of the whole population of trials. The Cochrane Collaboration has been established to minimize biases in reviews and meta-analysis. The first Cochrane Centre on Schizophrenia has been established in Oxford and others are planned.