Customers' choice decisions often involve two stages during which customers first use noncompensatory rules to form a consideration set and then make the final choice through careful compensatory tradeoffs. In this work, we propose a two-stage network-based modeling approach to study customers' consideration and choice behaviors in a separate but integrated manner. The first stage models customer preferences in forming a consideration set of multiple alternatives, and the second stage models customers' choice preference given individuals' consideration sets. Specifically, bipartite exponential random graph (ERG) models are used in both stages to capture customers' interdependent choices. For comparison, we also model customers' choice decisions when consideration set information is not available. Using data from the 2013 China auto market, our results suggest that exogenous attributes (i.e., car attributes, customer demographics, and perceived satisfaction ratings) and the endogenous network structural factor (i.e., vehicle popularity) significantly influence customers' decisions. Moreover, our results highlight the differences between customer preferences in the consideration stage and the purchase stage. To the authors' knowledge, this is the first attempt of developing a two-stage network-based approach to analytically model customers' consideration and purchase decisions, respectively. Second, this work further demonstrates the benefits of the network approach versus traditional logistic regressions for modeling customer preferences. In particular, network approaches are effective for modeling the inherent interdependencies underlying customers' decision-making processes. The insights drawn from this study have general implications for the choice modeling in engineering design.