This paper describes an experimental system which provides heuristic solutions to vehicle routing problems. The system basically contains two components: an initial routes generator and a route improver. The route generator consists of an modified version of the sweep algorithm, and a new TSP algorithm, called the shrink algorithm. It is observed that the shrink algorithm is much more computationally efficient than Lin's exchanged method, while giving outputs comparable to those from the latter. The route improver in turn consists of two sub-components. The first is a small yet efficient expert system, which identifies and remedies specific problems in the initial routes, with the experience of human routers incorporated in the improvement process. The second sub-component is a pair-wise re-router, which runs the saving algorithm on neighboring-route pairs, trying to reduce the total costs. The system has been tried on several typical routing problems, and yields optimal or near-optimal solutions.