The demonstration that 2 disorders coaggregate in families is often the first step in the exploration of genetic factors common to the 2 disorders. Previous methods of analyzing familial coaggregation have used either (1) a typical measure of association (eg, the odds ratio) between a disorder in an individual and another disorder in family members, or (2) a linear structural equation model (SEM). The association method accommodates case-control sampling of families, but may not assess the causal effect of interest because it is not based on an underlying causal model. The SEM method is based on a causal model, but cannot easily accommodate case-control sampling or direct effects of 1 disorder on the other within individuals. We develop a new method of analyzing coaggregation based on directed acyclic graphs. Because this method is a generalization of structural equation models and uses measures of association that accommodate case-control sampling and direct effects, it combines the strengths of both previous methods. In the absence of direct effects between disorders, our approach provides a valid estimate of the causal coaggregation effect. In the presence of direct effects, our approach provides an upper-bound estimate and (assuming additive linear effects of latent familial and nonfamilial factors) a lower-bound estimate of the causal coaggregation effect. For illustration, we applied our method to a family study of binge eating disorder and bipolar disorder.