OBJECTIVE The objective of the study was to examine the efficacy and the degree of adverse effects connected with atypical neuroleptic drugs and haloperidol by using a previously described Bayesian statistical method that includes both direct and indirect comparisons simultaneously. METHODS The authors used the results of 30 double-blind, randomized studies including comparisons of 4 atypical neuroleptics and haloperidol, head-to-head or against placebo. We calculated the response ratios for drugs against placebo and thereafter the relative response ratios for one drug against another. With uniform priors, we calculated and ranked the posterior estimates of response ratios for antipsychotic effect, weight gain, and occurrence of extrapyramidal symptoms. RESULTS All second-generation neuroleptics analyzed are fairly effective with response ratios against placebo ranging between 1.55 (credibility interval, 1.36-1.76) and 1.99 (1.76-2.26), with clozapine being the most effective and aripiprazole the least effective among them. The risk of inducing weight gain is clearly very high for all 5 neuroleptic drugs compared with placebo with response ratios of 12.21 (10.22-15.05) for olanzapine and 11.28 (6.89-17.77) for clozapine. There is a clear increased risk of extrapyramidal adverse effects for haloperidol compared with placebo as the response ratio is 2.33 (2.03-2.49). The other drugs all have considerably less risk of extrapyramidal adverse effects. CONCLUSIONS The 4 second-generation neuroleptics included in our meta-analysis show only small differences in overall efficacy, with clozapine being the most effective and aripiprazole the least effective among them. When the risk of adverse effects is analyzed, olanzapine and clozapine are afflicted with the highest risk of inducing weight gain and haloperidol with extrapyramidal symptoms. Even aripiprazole and risperidone, however, induce considerable weight gain compared with placebo but may be acceptable alternatives when tailoring drug treatment to the individual patient.