Bipolar affective disorder (BPAD) is a multifactorial disorder with various clinical presentations. Etiologic heterogeneity may partly underlie the phenotypic heterogeneity. Efforts to dissect BPAD have been based on the course of the disorders (BP I versus BP II or rapid cycling), comorbidity pattern (panic attacks, suicide attempts, addiction or hyperactivity), differences between the sexes, and clinical pattern (cycloid and puerperal psychosis). The present article provides a comprehensive review of the existing data, showing that age at onset (AAO) identifies homogeneous sub-groups of patients with BPAD. Recent work has demonstrated the existence of three--early, intermediate and late--onset bipolar sub-groups based on AAO, following Kendell's criteria for validity (The American Journal of Psychiatry 2003; 160: 999). We will also show how these distinctions may be of use in the search for genetic vulnerability factors and other pathogenic influences. Following Kendell's criteria, we show that AAO of bipolar disorders has been tested with most of the available strategies for establishing the validity of clinical syndromes. We also present data from genetic epidemiologic studies in bipolar disorder, showing that AAO sub-groups may reduce the underlying genetic heterogeneity. No accurate AAO thresholds to define valid sub-groups have been identified precisely. Until recently, studies defined early- and late-onset as corresponding to early or mid-adulthood, not taking into account juvenile-onset bipolar disorder. A recently proposed theoretical model with three AAO sub-groups (onset age 17, 27 and 46) is discussed.