Network representation is required to be simple and to have a high affinity to observed data, considering large-scale transportation network analysis. With the spread of technologies such as probe vehicles, continuous acquisition of detailed traffic data in a large-scale network is now possible. It is needed to link characteristic values to each link of network data for utilizing that. However, handling the data linked to all links of a detailed network can be very difficult when the number of links in the network is very large. In that case, aggregating a network structure is an effective approach, however, existing methods have some issues regarding the subjectivity of network selection or the dependence on the original network structure. In this paper, we developed a method to generate an aggregated network consisting of observed vehicle trajectories. Using observed vehicle trajectories to represent network can improve the objectivity of network representation and relieve the dependence on the original network data. As shown by numerical examples of Kobe area network, the complexity of the structure of the aggregated network is not too simple to lose information under network-wide traffic conditions and not too complex to incur a huge calculating cost.