A paucity of naturalistic data supported a rationale for the present retrospective review of clinical changes during hospitalization in 44 bipolar pre-adolescents, treated with monotherapy lithium, carbamazepine (CBZ) or divalproex sodium (DVP). Daily staff progress notes and discharge summaries on each patient were read by four trained clinicians blind to treatment group, and rated according to the Clinical Global Impression Improvement (CGI-I) scale. Consensus rating was measured by kappa reliability. Data were analyzed using a general linear model (sas mixed) analysis of variance (anova) with repeated measures. The medication groups did not differ in length of hospitalization, overall severity of illness at the time of admission, or comorbidity. Prior treatment was considered as a covariate. Each group approached serum therapeutic levels at day 7 of the medication period. The estimated mean CGI-I scores for CBZ were systematically higher (i.e. worse) than those for lithium and DVP, which overlapped and crossed over time. The difference became increasingly apparent and was statistically significant by week 2 (P = 0.036). The present study was limited in that the sample sizes, particularly in the case of CBZ, were small, commensurate with the low prevalence of the disorder. Lack of structured interviews, as an independent assessment of diagnoses was an intrinsic limitation of the study. Although constrained by its retrospective nature, our findings suggest that by week 2 of hospitalization both lithium and DVP may be more efficacious than CBZ in bipolar pre-adolescents. Any significant finding must be viewed as tentative and subject to confirmation in other studies.