A mathematical approach is developed for solving a new vehicle routing problem with backhauling (VRPB). The model is capable of dealing with multiple vehicles and multiple depots, both of which are capacitated. Because of the computational complexity inherent in solving the VRPB, the problem is decomposed into submodels. The procedure consists of 3 phases: 1. allocation of customers and vendors to capacitated clusters, 2. assignment of customers and vendors to depots and routes, and 3. individual route configuration. The decomposition procedure allows the solving of a real-world problem with 3 depots, 134 customers, and 27 vendors in 130 CPU seconds on a VAX-11/780 system. The capacitated depot solution consisted of a total of 15 tours from 3 depots. Twelve of the tours included both deliveries and pickups (backhauls), 2 included deliveries only, and one consisted solely of pickups.