Wind energy plays a significant role in renewable energies. The increasing demand for wind energy has caused wind turbines (WT) to grow steadily larger, which means that the control objectives are no longer solely to maximize the energy produced but to control mechanical loads, among other objectives actively. Model-based WT control, particularly model predictive control (MPC), has been the focus of research for the last decades. Nevertheless, only a few practical investigations of MPC for WTs in field tests exist.This paper highlights some key challenges and pitfalls when applying MPC for WTs. We render these critical points based on the experience of a recently conducted field test and discuss possible solutions for these challenges. In doing so, we highlight the following three critical areas: Firstly, we show how the design and practical operation of an MPC system can take into account the nonlinear properties of the WT. In particular, we address the highly varying sensitivity to the pitch angle and the dynamic responses of the rotor speed and mechanical loads to the actuator commands over the partial and full load ranges. Secondly, we discuss the problem of having limited computational capacities on real-time platforms, restricting the possible complexity of the MPC algorithm. Lastly, we show how some safety aspects decisively influence the design and operation of the control algorithm.