Wind loads encountered by wind turbine systems exhibit obvious randomness. Conventional researches of wind loads mainly target developing power spectrum density models or correlation functions. However, it is extremely difficult to deal with several key problems of stochastic dynamics of wind turbine systems in such a frame due to the absence of physical mechanisms. In this paper, the physical tradition in the study of stochastic dynamics is followed to investigate wind loads of wind turbine systems. Along the physical tradition, the physics-based modeling is primarily put forward. Based on the random function description of stochastic processes, a new concept of random Fourier function is proposed to establish two physical wind spectrum models of wind turbine systems. One of them is the random Fourier spectrum model which determines wind loads on wind turbine towers, the other is the rotational Fourier spectrum model which decides wind loads on rotating blades. As the two physical wind spectrum models can effectively prevent the essential drawbacks of the phenomenological ones, they are suggested to be adopted in the wind-induced dynamic reliability evaluation of wind turbine systems.