The notion that group- or community-level factors may play an important causal role in the health of individuals has received increasing attention in recent years. This interest reflects growing recognition that health variations across individuals cannot be fully explained exclusively as a function of individual-level characteristics. Social capital, which Fujiwara and Kawachi 1 define as the “resources accessed by individuals and groups within a social structure that facilitate cooperation, collective action, and the maintenance of norms” is one construct that has been hypothesized to affect health. By definition, social capital refers to social relationships and connections between individuals and is therefore a relational rather than a purely individual-level attribute. A number of studies have reported associations between measures of social capital for geographically defined contexts (such as states or neighborhoods) and a variety of health outcomes after statistical controls for individual-level characteristics. 2 However, as often noted, observational studies have important limitations in their ability to control for confounders. Fujiwara and Kawachi 1 use an elegant twin design to examine whether differences in physical and mental health between twins are linked to differences in their reports of social capital. By focusing on within‐twin pair comparisons, this design controls for shared early life environments (to the extent that twins reared together share the same environment) and for shared genetic factors (all genetic factors in monozygotic twins and some genetic factors in dizygotic twins). The main finding reported by Fujiwara and Kawachi 1 is that within both monozygotic and dizygotic twin pairs, the twin reporting greater social trust in his or her neighborhood (as indicated by agreement with the statement people in my neighborhood trust each other) also tended to report better physical health. This is an interesting finding and an improvement over prior observational studies in that it inherently controls for many other factors that twins share and which may be related to physical health and reports of social trust. A limitation, however, is that it does not necessarily control for all lifecourse and adult factors on which twins may differ. Some of these (such as education and very broad employment categories) were statistically controlled in the analyses reported by Fujiwara and Kawachi, but to skeptics the possibility of residual confounding remains. Another limitation noted by the authors is that because both social trust and health are self-reported, same-source bias (or common-method bias) remains a possibility. For example, certain individuals may be more pessimistic in their reports regarding themselves and their neighborhoods, generating spurious associations between both variables. The availability of objective measures of health (through measured outcomes) and social trust (through raters or responses of neighbors) would eliminate this problem. The possibility of reverse causation is a third limitation: This cross-sectional analysis cannot rule out the possibility that ill health causes people to see their neighborhoods in a more negative light.