This work investigates design and implementation of a Central Pattern Generator as loco- motion controller for a modular robot. In nature, locomotion is an important criteria for the survival of a species: it is needed to reach food, escape enemies, move to a friendlier environment and enables sexual reproduction. For human and animal organisms this important skill, as well as many more, is enabled by Central Pattern Generators (CPGs). CPGs are present as distributed neural networks in the spinal cord and transform excitatory input from the brain stem into oscillating control signals for the muscles. This oscillation is generated without the need for an oscillating input. Artificial CPGs have been imple- mented on robots for a while, but there are not many works regarding modular robotic organisms. This work compares previously developed CPG models, both connectionist an non-connectionist ones, and a custom model, which constitutes a hybrid between both, is developed. Locomotion is evolved using Evolutionary Programming for different organism shapes, and sensory feedback as well as a distributed implementation and high level control mechanisms for the controllers are taken into account. Evolution of the CPG controller is realized in a simulated environment for a caterpillar-like, a quadruped and a randomized body shape. The result of this work is a CPG controller which is capable of evolving loco- motion, providing low-level reflexes to react on obstacles and changes in the environment and adjusting to arbitrary body shapes of the modular robot.