<p>Urban road networks form the backbone of cities, shaping mobility, accessibility, spatial organization and urban form. However, widely used road-network data sources, such as OpenStreetMap, are primarily organized for general geographic representation and navigation-oriented applications, and they often require further processing before they can support systematic urban studies. In particular, existing datasets rarely provide standardized multi-representational versions of the same road networks, although geometric, topological, semantic, and cognitive perspectives are all essential for understanding different aspects of urban structure. Here, we present a dataset of multiple structural representations of urban road maps, covering 293 prefecture-level cities and four provincial-level municipalities in China. The dataset provides simplified and topologically reconstructed road networks in four complementary graph models: segment-based primal graphs, segment-based dual graphs, stroke-based dual graphs, and mixed dual graphs that integrate named streets with strokes. Together, these representations capture the geometric, topological, semantic, and cognitive properties of urban road networks within a standardized framework.Validation demonstrates high geometric fidelity (R² = 0.991) and strong topological consistency, with 99% of original connectivity retained after processing. Provided in GIS-compatible Shapefile and GraphML formats, the dataset offers an open-access and reproducible foundation for research on urban morphology, spatial cognition, accessibility, space syntax, and complex network analysis. It is designed primarily for morphological and structural analysis of urban road networks, rather than detailed operational traffic modelling requiring lane capacity, turning restrictions, signalisation, travel speeds, real-time traffic conditions, or fine-grained directionality.</p>