For vehicle routing problems (VRP) with time windows (VRPTW) solved by conventional cluster-first and route-second approach, temporal information is usually considered with vehicle routing but ignored in the process of clustering. The authors propose an alternative approach based on spatiotemporal partitioning to solving a large-scale VRPTW, considering jointly the temporal and spatial information for vehicle routing. A spatiotemporal representation for the VRPTW is presented that measures the spatiotemporal distance between two customers. The resulting formulation is then solved by a genetic algorithm developed for k-medoid clustering of large-scale customers based on the spatiotemporal distance. The proposed approach showed promise in handling large scale networks.