This paper presents a method for solving vehicle routing problems (VRPs) using a local search technique called Large Neighbourhood Search (LNS). LNS is inspired by the shuffling technique used in job-shop scheduling and integrates well with constraint programming. The method involves removing a set of customer visits from the current solution and re-inserting them using a constraint-based tree search. Limited Discrepancy Search is used during the re-insertion process to improve the solution. The method is tested on benchmark problems and shows competitive results compared to operations research meta-heuristics, indicating that constraint-based technology is applicable to VRPs. The paper introduces LNS as a technique for VRPs, which involves removing and re-inserting customer visits. The method uses a "relatedness" measure to select which customer visits to remove and re-insert. The size of the set of customer visits removed increases over time when the search stagnates. The re-insertion process uses heuristics and constraint propagation to evaluate the cost and legality of the move. The paper compares LNS with related work and assesses its benefits. It presents computational experiments on benchmark problems, both with and without time windows. The results show that LNS has excellent average performance and produces many new best solutions to these benchmark problems. The paper concludes that LNS is a promising approach for solving VRPs, combining the advantages of local search and constraint programming.This paper presents a method for solving vehicle routing problems (VRPs) using a local search technique called Large Neighbourhood Search (LNS). LNS is inspired by the shuffling technique used in job-shop scheduling and integrates well with constraint programming. The method involves removing a set of customer visits from the current solution and re-inserting them using a constraint-based tree search. Limited Discrepancy Search is used during the re-insertion process to improve the solution. The method is tested on benchmark problems and shows competitive results compared to operations research meta-heuristics, indicating that constraint-based technology is applicable to VRPs. The paper introduces LNS as a technique for VRPs, which involves removing and re-inserting customer visits. The method uses a "relatedness" measure to select which customer visits to remove and re-insert. The size of the set of customer visits removed increases over time when the search stagnates. The re-insertion process uses heuristics and constraint propagation to evaluate the cost and legality of the move. The paper compares LNS with related work and assesses its benefits. It presents computational experiments on benchmark problems, both with and without time windows. The results show that LNS has excellent average performance and produces many new best solutions to these benchmark problems. The paper concludes that LNS is a promising approach for solving VRPs, combining the advantages of local search and constraint programming.