We construct a novel structural model for the purposes of testing for the presence of spillover effects from social capital. There is a disconnect in the social science literature as to whether social capital is an attribute of individuals (as typically modeled in economics and some sociology work) or an attribute of communities (as in the political science, sociology and public health disciplines). However, existing empirical work on coimmunity social capital simply uses aggregations of individual attributes to measure community level social capital (e.g., Putnam 20000); such measures may simply proxy for unobserved individual attributes. Consequently, there is currently no evidence that social capital is an attribute of communities (or groups). We develop a strutural model that allows us to test whether individual trust and membership in voluntary organizations are determined by community level trust and membership (ie, are there spillovers from social capital?). We estimate these equations simultaneously in a SUR framework; we also instrument for community level trust and membership using other community level attributes. Finally, because we are working with mmultilevel data, we correct for the clustering of observations at the community level. This work represents not only a major advance in the methods applied to social capital research, but also the first systematic evidence that community level social capital spills over and influences individual level social capital. The substantive importance of this finding is that it generates plausible instruments for identifying teh treatment effect of social capital on various outcome variables of interest, including voter turnout, trust in government, employment, and health status. [ABSTRACT FROM AUTHOR]