While the literature of social networks and entrepreneurship has stated the importance of social networks for entrepreneurship, many past empirical studies tended to measure social networks only at the aggregated level. This led to a failure to capture heterogeneity or nonredundancy of networks, and sometimes resulted in empirically confusing verdicts regarding the contribution of social networks to entrepreneurship at a statistical level. We argue that disaggregation of networks is critical and will construct networks at a refined, specific level. With a large-scale dataset of 2,273 individuals, we test the link between start-up activities and Chinese and Indian ethnic professional associations in Silicon Valley. Our multivariate models find that only a few networks with heterogeneous members correlate positively with entrepreneurship. The results indicate that social networks, a powerful and widely used theoretical construct, are still underdeveloped in both conceptualization and operationalization, and that we have to further develop effective measurement techniques.