OBJECTIVES: In a direct comparison of patients with bipolar manic or mixed episodes, olanzapine and risperidone showed similar efficacy in mania ratings (Baker, APA 2003), but olanzapine had greater efficacy on clinician global impression of severity (CGI-S). We performed a post-hoc analysis investigating possible determinants of this differential treatment effect. METHODS: This 3-week, double-blind study compared olanzapine (5-20 mg/day; N = 165) to risperidone (1-6 mg/day; N = 164) in manic or mixed episodes. Path analysis dissected the CGI-S treatment effect into drug effect explainable by effects on manic, depressive and extrapyramidal symptoms, measured by standard rating scales, versus other treatment effects not accounted for by rating scales. Demographic and disease characteristics were examined to determine if they were significant predictors of CGI-S improvement. Analysis of variance and regression models used change from baseline to endpoint (last observation carried forward) for all analyses. RESULTS: Olanzapine-treated patients achieved significantly greater improvement in CGI-S than risperidone-treated patients (p = 0.014). While changes in manic and depressive symptoms were significant predictors of the change in CGI-S (both p <0.001), treatment effect remained statistically significant (p = 0.006) even after adjustment for change in these effects. In fact, 93% of olanzapine's superior treatment effect on CGI-S was not attributable to changes in mania, depression, or extrapyramidal symptoms. Race, gender, rapid cycling status, manic vs. mixed diagnosis, age of disease onset, and weight change were not significant determinants of CGI-S change. A significant interaction was detected between age and treatment: older olanzapine-treated patients had greater CGI improvement compared to older risperidone patients; no differences occurred between treatments for younger patients. CONCLUSIONS: Rating scale, demographic, and illness characteristics were at most modest determinants of improvement in global illness severity. Even after adjusting for significant determinants, olanzapine-treated patients still experienced significantly greater improvement than risperidone-treated patients in global illness severity.