Practice should, as much as possible, be based on good science. Randomized clinical trials can provide the best evidence, but they have serious limitations. First, many clinical situations, such as cardiac arrest and pain relief, do not lend themselves to randomization. Second, trials seldom can study the effects seen in different subgroups, nor can the results always be extrapolated from the restricted groups of patients recruited into trials. Finally, there is publication bias: the failure to report "negative" trials and the biased presentation of results by investigators and sponsors.