Electronic map data is of enormous importance for analyses of travel, urban congestion, etc. It is growing an important basic data of transportation research and widely employed in vehicle navigation system and intelligent transportation system. Of electronic map data, road network data is a fundamental part and often updated. As a result of a lot of expense of manpower and resources of traditional method, more and more research efforts have been made on extracting geometrical characteristics of road network from other data resources like LIDAR data. The wide use of mobile positioning devices makes it possible to obtain a large volume of vehicle trajectory data within a short period. These data are complete records of vehicles' movement and able to reflect geometrical characteristics of road network as well. By this means, vehicle trajectory data can be employed to figure out changes happening to the spatial extent of road networks as long as enough data are available, and then we can update the road network database. Since trajectory data integrate both spatial and temporal information and their volume is often huge, it is not straightforward to realize the aforementioned target. In view of this, the paper proposes a thinning-algorithm-based method to construct road network. Vehicle trajectory data are vector data, are first converted to raster data, then thinning method can be used to extract geometrical characteristics of road network. With this method, center lines of road networks can be extracted, network topology be maintained, and meanwhile redundant information be removed. A case study in Lujiazui, Shanghai with taxi trajectory data is conducted and the results are evaluated.