Experiments were conducted to investigate the way humans solve Capacitated Vehicle Routing Problems (CVRPs), a problem class in which the shortest set of tours must be found around a set of weighted nodes using a capacity-limited vehicle. The first two experiments explored human performance in drawing solutions to problems of different complexity in terms of number of routes, nodes and weights to be summed. They also included as an experimental factor Verbalisation, both to provide a qualitative indicator of performance and also to examine the impact of verbalisation on performance. The qualitative results of Experiment 1 indicated two major types of strategists: Calculators and Clusterers. Clusterers performed faster and in some of the problems found solutions closer to the optimal than calculators. The major errors that participants performed were errors of calculation, nodes missing and drawing too few routes. Results from Experiment 2 suggest that humans are showing the best performance in problems with low calculation demands while they exhibit the worst performance in the problems with negligible calculation demands, thus suggesting that in order to provide very close to optimal solutions in CVRPs it is necessary to retain some calculation demand load to promote a more optimising behaviour. New strategies have been revealed in Experiment 2 and Verbalisation again did not influence the human performance. Further qualitative and quantitative analyses of the verbalisations and human performance in Experiment 1 showed that Visuospatial strategies such as Anchoring and Clustering are predictors of good performance while Arithmetic strategies such as Balancing generate poor performance. In Experiment 2, the best performances were exhibited when participants were using either Visuospatial strategies or Arithmetic strategies. The success and failure of the adoption of these strategies is dependant on the problem complexity and the cognitive load. A third 14 ..... ------------.~~-~~~ -- experiment revealed that error-trapping did not influence the human performance. The results informed the specification and design of a Capacitated Vehicle Routing Problem Solver implemented in Java. A pilot study was completed that led to a revaluation of the software. A later version was implemented and tested empirically. Experiment 4 revealed that humans interaction with the Capacitated Vehicle Routing Problem Solver to solve CVRPs significantly improved their performance leading to the generation of very close to optimal routes.