Recent publications on automatic flight controls are concentrated on neural network based adaptive controls. The ability of the neural networks to approximate the continuous smooth functions has been explored for adaptation. In this report the design details of an adaptive controller for an unmanned helicopter using these modern concepts is explained. This controller consists of feedback linearization, linear model inversion and learning while controlling neural network architecture. The time scale separation between position and attitude dynamics of the vehicle is used in the controller synthesis. In order to simplify the nonlinear controller, approximations to the body axis forces are used in the controller calculation. The attitude control uses the neural networks to adaptively cancel the inversion errors. Finally the design is verified using the simulation results.