In the next decades wind energy is expected to secure a firm share of the total energy production capacity in many countries. To increase competitiveness of wind power with other power sources it is essential to lower the cost of wind energy. Given the design of a turbine, this objective can be attained in several ways, for instance by increasing the energy production of a wind turbine and by lowering loads on the wind turbine in order to reduce maintenance costs. Research performed in recent years has shown that feedback control plays an important role in these two aspects. Refined control design can increase power production, for instance by using feedforward information about the wind field. At the same time, control can reduce wear of the turbine by mitigating fatigue and extreme loads. For the design process of new and advanced control concepts which meet these objectives, detailed models are essential. System identification on the basis of experimental data can provide such models and help to understand differences between the behaviour of theoretical models and the real wind turbine. In this thesis we address specific challenges related to the application of system identification techniques to wind turbines and similar systems. The result is a set of methods which can be applied to wind turbines in practical situations.