Bipolar disorder is one of the most costly psychiatric disorders to treat. OBJECTIVE: This analysis focuses on (direct) cost outcomes of a multi-center, randomized, double-blind clinical trial comparing efficacy, safety, effectiveness, and costs of two first-line treatments in acute mania, olanzapine versus divalproex. METHODS: The original study included 251 inpatients with a diagnosis of bipolar I disorder, in an acute manic or mixed episode. Hospitalization was required for the first week of double-blind treatment, with subsequent discharge if clinically appropriate. Inpatient and outpatient resource utilization data were collected at the end of the acute phase of the trial (3 weeks), and during the maintenance phase (weeks 7, 15, 23, 31, 39, and 47). Costs were analyzed for all patients who entered the maintenance phase (olanzapine n=77, divalproex n=70). Acute and maintenance phase costs were estimated by assigning prices (in year 2000 dollars) from a standard list to units of applicable medical services. Using a trimmed-means (25% on each tail of the cost distribution) comparison to overcome the skewness in the distribution of cost data, the medication treatment groups were compared. RESULTS: Overall per-patient costs were not significantly different between the olanzapine-treated patients (15.9 ± 4.5 mg/day) and the divalproex-treated patients (1596.4 ± 492.7 mg/day). However, olanzapine treatment was associated with significantly higher medication costs (p < .001), but significantly lower outpatient (p < .001) and overall inpatient (p < .05) costs over the course of treatment. Outpatient costs were higher in divalproex-treated patients due to higher emergency room and other outpatient visits. CONCLUSIONS: These findings suggest that differences in medication acquisition cost are offset by lower costs for other clinical services during olanzapine treatment. Further research is needed to determine the extent to which the present findings can be generalized to practice settings outside of the clinical trial context.