Health services often manage agitated or violent people and for emergency psychiatric services such behaviour is particularly prevalent (10%). The drugs used in this situation should ensure that the person swiftly and safely becomes calm.To examine whether haloperidol plus promethazine is an effective treatment for psychosis induced agitation/aggression.We searched the Cochrane Schizophrenia Group's Register (July 2004).We included all randomised clinical trials involving aggressive people with psychosis for which haloperidol plus promethazine was being used.We reliably selected, quality assessed and extracted data from all relevant studies. For binary outcomes we calculated standard estimations of risk ratio (RR) and their 95% confidence intervals (CI). Where possible we estimated weighted number needed to treat or harm (NNT/H).We identified two relevant high quality studies. One compared the haloperidol plus promethazine mix with midazolam (n=301) and one with lorazepam (n=200). The combined results were largely heterogeneous. In Brazil, haloperidol plus promethazine was an effective means of tranquillisation with over two thirds of people being tranquil or sedated by 30 minutes, but midazolam was more swift (n=301, RR 2.9 CI 1.75 to 4.80, NNH 5 CI 3 to 12). In India, however, 95% of people were tranquil or sedated by 30 minutes if allocated to the combination treatment (vs lorazepam, n=200, RR 0.26 CI 0.10 to 0.68, NNT 8 CI 6 to 17). Over the next few hours of treatment reported differences are negligible. One person given midazolam had respiratory depression (reversed by flumazenil), one given lorazepam had respiratory difficulty. A single person given haloperidol plus promethazine had an epileptic fit. Once the initial tranquillisation was administered, few needed additional medications for continued agitation (n=501, 2 RCTs, RR needing additional tranquillising drugs by four hours 1.67 CI 0.62 to 4.54, 4% vs 2%, I squared 50%) and there were no differences in the low levels of use of restraints. About 28% of people in Brazil in both groups had another episode of aggression in the first day after the initial injection (n=301, RR 0.89 CI 0.62 to 1.29). About half of all people in the Indian study were discharged by four hours (n=200, RR 1.13 CI 0.85 to 1.50) and a similar proportion in Brazil by 15 days (n=301, RR 1.05 CI 0.84 to 1.29). Both studies attained 99% follow up for their primary outcomes. Even by two weeks only 4% of people could not be accounted for (n=501, 2 RCTs, RR 0.91 CI 0.38 to 2.17).This review suggests that both benzodiazepines work, but that midazolam has a faster onset and thereby reduces the risk of exposure to violence. Both benzodiazepines have the potential to cause respiratory depression, probably midazolam more so than lorazepam, and we would question the use of this group of drugs outside of those services fully confident of observing for and managing the consequences of respiratory distress. Most evidence, however, exists for the haloperidol plus promethazine mix, with currently more than 400 people randomised to the combination. The onset of action is swift and faster than lorazepam. The combination also seems safe with no clear longer term consequences. We would expect policy makers recommending other drug managements to have equally compelling evidence to support their guidance and hope that this would not be founded in conjecture or consensus, which may be more difficult to defend than evidence from high quality studies.