Background: The array of different diagnoses and clinical presentations seen in the family members of bipolar probands suggests a quantitative or spectrum phenotype. Consistent with this idea, it has been proposed that an underlying quantitative variation in temperament may be the primary phenotype that is genetically transmitted and that it in turn predisposes to bipolar disorder (BP). Choosing the appropriate phenotypic model for BP is crucial for success in genetic mapping studies. To test this theory, various measures of temperament were examined in the family members of bipolar probands. We predicted that a gradient of scores would be observed from those with BP to those with major depression to Unaffected relatives to controls. Methods: Members of 85 bipolar families and 63 control subjects were administered clinical interviews for diagnosis (SCID) and two temperament assessments, the TEMPS-A and TCI-125. Subjects with BP, major depressive disorder. unaffected relatives, and controls were compared on each temperament scale and on eight factors extracted from a joint factor analysis of the TEMPS-A and TCI-125. Results: The four groups were found to be significantly different and with the expected order of average group scores for four of the TEMPS-A scales, three of the TCI-125 scales, and one of the extracted factors. On the fifth TEMPS-A scale, hyperthymic, controls scored higher than the other three subject groups contrary to expectations. Significant differences were seen between unaffected relatives and controls on the hyperthymic scale and on the first extracted factor, anxious/reactive. Limitations: Controls were mainly recruited through advertisements, which may have introduced an ascertainment bias. It is also possible that mood state at the time of completing the questionnaire influenced subject's rating of their temperament. Additionally, bipolar I and bipolar II subjects were placed in the same group even though they had some differing clinical features. Conclusions: Our data support the theory that some dimensions of temperament are transmitted in families as quantitative traits that are part of a broader bipolar spectrum. In particular, the hyperthymic scale of the TEMPS-A and the anxious/reactive extracted factor distinguished unaffected relatives from controls. The hyperthymic scale yielded results opposite to expectation with controls higher than any family group. This may be an artifact of the self-rated form of the questionnaire, a consequence of our grouping bipolar I and II subjects together, or the result of a "protective" factor and bears further study. Nevertheless, both of these scales may be useful quantitative traits for genetic mapping studies. (c) 2003 Elsevier B.V. All rights reserved.