This paper introduces a neural network based model reference adaptive control architecture that allows adaptation in the presence of saturation. The given plant is approximately feedback linearized, with adaptation used to cancel any matched uncertainty. A nested saturation based reference model is used. This law allows the incorporation of magnitude actuator saturation and has useful small gain properties. Depending on the bandwidth and saturation limits, the reference model based on this law eases off on the aggressiveness of the desired trajectory thus avoiding saturation. However, actuator saturation might yet occur due to uncertainty or external disturbances. In order to protect the adaptive element from such plant input characteristics, the nested saturation reference model is augmented with a pseduo-control hedging signal that removes these characteristics from the adaptive element's training signal.