Many studies have used admixture analysis to separate age-at-onset (AAO) subgroups in bipolar patients, but few have looked at the phenomenological characteristics of these subgroups, in order to find out phenotypic markers. Admixture analysis was applied to identify the model best fitting the observed AAO distribution of a sample of 1082 consecutive DSM-IV bipolar I manic inpatients who were assessed for demographic, clinical, course of illness, comorbidity, and temperamental characteristics. The model best fitting the observed distribution of AAO was a mixture of three Gaussian distributions. We could identify three AAO subgroups: early, intermediate, and late age-at-onset (EAO, IAO, and LAO, respectively). Patients in the EAO subgroup were more often single young males exhibiting severe mania with psychotic features, a subcontinuous course of illness with substance use and panic comorbidity, more suicide attempts, and temperamental components sharing hypomanic features. Patients with LAO showed a less severe picture with more depressive temperamental components, alcohol use and comorbid general medical conditions. A less typical phenotype was present in IAO patients. The following are the limitations of this study: retrospective design, and bias toward preferential enrollment of patients with manic predominant polarity. This study confirms that bipolar I disorder can be subdivided into three subgroups based on AAO distribution and shows that patients from these subgroups differ in phenotypes. © 2013 Elsevier B.V. All rights reserved.