The research work in this thesis deals with both nonlinear system identification and anthropomorphic systems motion optimization. Firstly, we consider the identification of three systems that are: linear systems in multiple experiments case, finite degree Volterra series with infinite horizon and quadratic in-the-state systems. Thus we propose novel and efficient identification methods. Secondly, we apply the system identification techniques to synthesize and to model human locomotion. Besides, we address the optimization of humanoid robot motions, the imitation of human captured motions by a humanoid robot and finally the time parameterization of humanoid robot paths in the configuration space. The experimental validation of our proposed methods on the humanoid platform HRP-2 has pointed out that they not only outperform the conventional methods, but also provide a unified framework to control humanoid robots. In conclusion, the several contributions in this thesis provide new ways of thinking about some problems in nonlinear system identification and humanoid robot control.