In this study, we critically examine whether contextual social capital (CSC) is associated with self-rated health, with an emphasis on the problem of confounding. We also examine different components of CSC and their association with self-rated health. Finally, we look at differences in susceptibility between different socio-demographic groups. We use the cross-sectional base line study of the Stockholm Public Health Cohort, conducted in 2002. A postal questionnaire was answered by 31,182 randomly selected citizens, 18-84 years old, in Stockholm County. We used four measures of social capital: horizontal (civic trust and participation), vertical (political trust and participation), cognitive (civic and political trust) and structural (civic and political participation). CSC was measured at parish level from aggregated individual data, and multilevel regression procedures were employed. We show a twofold greater risk of poor self-rated health in areas with very low CSC compared with areas with very high CSC. Adjustments for individual socio-demographic factors, contextual economic factors and individual social capital lowered the excess risk. Simultaneous adjustment for all three forms of confounding further weakened the association and rendered it insignificant. Cognitive and structural social capital show relatively similar associations with self-rated health, while horizontal CSC seems to be more strongly related to self-rated health than vertical CSC. In conclusion, whether there is none or a moderate association between CSC and self-rated health, depends on the extent to which individual social capital is seen as a mediator or confounder. The association with self-rated health is similar independent of the measure of CSC used. It is also similar in different socio-demographic groups.