In this 4-week, double-blind, parallel group study, eight young women with primary anorexia nervosa were evaluated on lithium carbonate, and eight patients were treated with placebo and served as a control. All patients participated in a behavior modification treatment program. The lithium-treated and placebo groups were comparable on nearly all findings measured at baseline (t tests), with no significant differences observed except for calories per day, percent fat composition of the daily calories, "interpersonal sensitivity" on the Hopkins Symptom Checklist-90 (HSCL-90), "self-care" on the Goldberg Anorectic Attitude Questionnaires, (GAAQ) and "manipulation of others" on the physician-rated Psychiatric Rating Scale (PRS). The data were analyzed using repeated measures analysis of covariance (ANCOVA) with the baseline measure as the covariate. Group differences appeared in the areas of "denial or minimization of illness" on the GAAQ, "selective appetite" on the PRS, and weight. Although the repeated measures ANCOVA for weight revealed a significant group-by-time interaction, indicating nonparallelism and invalidating the test for group differences, ANCOVAs performed for each individual time point showed greater weight gain in the lithium group at weeks 3 and 4.