A quadrotor helicopter is a highly nonlinear and open-loop unstable system, presenting significant challenges in terms of control design. This thesis proposes a neural network based controller, which makes use of a new robust technique to avoid weight drift caused by persistent disturbances, and describes the development of a working prototype for testing purposes. The controller consists of an adaptive neural network for attitude/ altitude control and adaptation, and a position control to design roll and pitch angles/velocities. When the new robust technique is applied in simulation, results demonstrate the control design is able to stabilize the quadrotor, while avoiding weight drift. After implementing the adaptive neural network on the prototype, experiments demonstrate that the control is able to stabilize the quadrotor, that nonlinear techniques provide a wider operating range than linear controls, and that weight drift is as much of a problem practically, as it is theoretically.