Job shop scheduling (JSS) problems are a group of optimisation problems which involves processing utilising a limited number of machines to process jobs as effectively as possible, where a machine can process one job at a time. Many manufacturing systems in real-world can be modelled as JSS problems. Approaches to JSS ranges from computationally intensive exact optimisation techniques, to dispatching rule heuristics that can generate good but not necessarily the best solutions to JSS problem instances. Hyperheuristics are high level heuristics which can automatically generate a smaller heuristic to a problem, and recent studies have applied hyper-heuristics to JSS problems. However, many of the hyper-heuristic approaches from the literature are limited in that they 1) only generate a single heuristic that are myopic in nature, 2) require a long computation time to generate a heuristic, 3) have mainly been applied to problems with no machine breakdowns, and 4) have no way of automatically updating a generated rule. The overall goal of this thesis is to develop a genetic programming based hyper-heuristic approach that aims to address the four limitations of existing hyper-heuristic approaches in the literature. This approach will combine novel dispatching rule representations with evolutionary computation mechanisms to meet all of these requirements.