Social contact networks represent proximity relationships between individual agents. Such networks are useful in diverse applications, including epidemiology, wireless networking and urban resilience. The vertices of a social contact network represent individual agents (e.g. people). Time varying edges represent time varying proximity relationship. The networks are relational -- node and edge labels represent important demographic, spatial and temporal attributes. Synthesizing social contact networks that span large urban regions is challenging for several reasons including: spatial, temporal and relational variety of data sources, noisy and incomplete data, and privacy and confidentiality requirements. Moreover, the synthesized networks differ due to the data and methods used to synthesize them. This dissertation undertakes a systematic study of synthesizing urban scale social contact networks within the specific application context of computational epidemiology. It is motivated by three important questions: (i) How does one construct a realistic social contact network that is adaptable to different levels of data availability? (ii) How does one compare different versions of the network for a given region, and what are appropriate metrics when comparing the relational networks? (iii) When does a network have adequate structural details for the specific application we have. We study these questions by synthesizing three social contact networks for Delhi, India. Our case study suggests that we can iteratively improve the quality of a network by adapting to the best data sources available within a framework. The networks differ by the data and the models used. We carry out detailed comparative analyses of the networks. The analysis has three components: (i) structure analysis that compares the structural properties of the networks, (ii) dynamics analysis that compares the epidemic dynamics on these networks and (iii) policy analysis that compares the efficacy of various interventions. We have proposed a framework to systematically analyze how details in networks impact epidemic dynamics over these networks. The results suggest that a combination of multi-level metrics instead of any individual one should be used to compare two networks. We further investigate the sensitivity of these models. The study reveals the details necessary for particular class of control policies. Our methods are entirely general and can be applied to other areas of network science. / Ph. D.