We sought to compare four different definitions of control groups in studies of the coaggregation between two disorders (A and B) on: 1) their ability to detect valid familial coaggregation; 2) their liability to artifactual evidence for familial coaggregation; and 3) their robustness to the overselection of comorbid cases. Using a quantitative genetic model of transmission, we simulated sibling pairs with familial and nonfamilial sources of comorbidity. Four different definitions of controls were tested to predict disorder B in siblings of cases vs. controls: 1) unscreened controls included subjects with A or B as well as subjects with either A or B; 2) in the symmetrical selection method, controls included only subjects without A; 3) supernormal controls included only subjects without A or B; and 4) in the pure proband method, cases included subjects with A only, and controls included only subjects without A or B. In the absence of selection bias, 1) the unscreened control and the symmetrical selection methods did not yield spurious evidence for familial coaggregation and could detect familial coaggregation; 2) the supernormal controls yielded spurious evidence of familial coaggregation; and 3) the pure proband method sometimes yielded spurious evidence for negative familial coaggregation, and had limited power to detect familial coaggregation. However, the pure proband method was the only one unaffected by overselection of comorbid cases. In the absence of selection bias, both the unscreened control and the symmetrical selection methods are appropriate, and the ro-bustness of the pure proband method to overselection of comorbid cases may be an interesting feature in studies using clinical samples. Moreover, quantitative genetics methods may offer important advantages in the study of familial coaggregation.