Reliable operation of helicopters in hover mode is essential for carrying out missions of surveillance, reconnaissance, and deployment of communication networks in disaster hit areas, among many others. Achieving autonomous operation in hover mode requires the development of robust model-based controllers. In this paper, the use of linear and nonlinear models to identify the orientation dynamics of a small scale helicopter is addressed. A linear architecture that combines the input-output dynamics and perturbation-output dynamics is introduced in this paper. In contrast to the linear models that have been reported in the literature, no assumptions about decoupled roll-pitch-yaw axes are made in the proposed approach. The nonlinear model of orientation dynamics is identified using artificial recurrent neural networks. Verification of these models is performed using actual data collected during the flight of the helicopter. The results show that incorporating the perturbation dynamics in the model can result in a description that can accurately predict the dynamics during actual flight conditions.