Research aimed at elucidating the underlying neurobiology and genetics of bipolar disorder, and factors associated with treatment response, have been limited by a heterogeneous clinical phenotype and lack of knowledge about its underlying diathesis. We used a survey of clinical, epidemiological, neurobiological, and genetic studies to select and evaluate candidate endophenotypes for bipolar disorder. Numerous findings regarding brain function, brain structure, and response to pharmacological challenge in bipolar patients and their relatives deserve further investigation. Candidate brain function endophenotypes include attention deficits, deficits in verbal learning and memory, cognitive deficits after tryptophan depletion, circadian rhythm instability, and dysmodulation of motivation and reward. We selected reduced anterior cingulate volume and early-onset white matter abnormalities as candidate brain structure endophenotypes. Symptom provocation endophenotypes might be based on bipolar patients' sensitivity to sleep deprivation, psychostimulants, and cholinergic drugs. Phenotypic heterogeneity is a major impediment to the elucidation of the neurobiology and genetics of bipolar disorder. We present a strategy constructed to improve the phenotypic definition of bipolar disorder by elucidating candidate endophenotypes. Studies to evaluate candidate endophenotypes with respect to specificity, heritability, temporal stability, and prevalence in unaffected relatives are encouraged.