Large-scale genome-wide association studies (GWASs) on bipolar disorder (BD) have implicated the involvement of the fatty acid desaturase (FADS) locus. These enzymes (FADS1 and FADS2) are involved in the metabolism of eicosapentaenoic acid (EPA) and docosahexaenoic acid (DHA), which are thought to potentially benefit patients with mood disorders. To model reductions in the activity of FADS1/2 affected by the susceptibility alleles, we generated mutant mice heterozygously lacking both Fads1/2 genes. We measured wheel-running activity over six months and observed bipolar swings in activity, including hyperactivity and hypoactivity. The hyperactivity episodes, in which activity was far above the norm, usually lasted half a day; mice manifested significantly shorter immobility times on the behavioral despair test performed during these episodes. The hypoactivity episodes, which lasted for several weeks, were accompanied by abnormal circadian rhythms and a marked decrease in wheel running, a spontaneous behavior associated with motivation and reward systems. We comprehensively examined lipid composition in the brain and found that levels of certain lipids were significantly altered between wild-type and the heterozygous mutant mice, but no changes were consistent with both sexes and either DHA or EPA was not altered. However, supplementation with DHA or a mixture of DHA and EPA prevented these episodic behavioral changes. Here we propose that heterozygous Fads1/2 knockout mice are a model of BD with robust constitutive, face, and predictive validity, as administration of the mood stabilizer lithium was also effective. This GWAS-based model helps to clarify how lipids and their metabolisms are involved in the pathogenesis and treatment of BD.