This paper presents results from comparison of an industry-accepted model of the western United States with two different synthetic network models of the same geographic area. The synthetic models were independently produced by different construction algorithms but based on public information concerning the location and size of generation and demand. New realism measures examined include Fiedler Vector Partitioning (FVP), the Spectrum of the Normalized Laplacian (SNL), Topological and Impedance Element Ranking (TIER) and Singular Value decomposition of the power flow Jacobian (SVJ). The synthetic models differ greatly in size and granularity, but each has similarities to the industry-accepted model. However, the industry-accepted model exhibits behaviour not replicated in either of the synthetic models. The suitability of each analysis to assess the realism of a network model is discussed.