Aeroelastic codes are fundamental for the design of wind turbines and the prediction of instabilities. Such codes rely on engineering models which limit their accuracy. With wind turbine blades becoming more slender to lower the costs, it is important to reduce these uncertainties to maintain a safe and stable turbine design. The dynamic stall model is one of these engineering models which simplifies the physics involved and so an improvement in accuracy might be possible. Various dynamic stall models have been published over the last 50 years. However, dynamic stall has proven to be a complex phenomenon to model accurately over a wide range of conditions and is an ongoing topic of research. Here four semi-empirical dynamic stall models, the Øye, Risø, Snel and ONERA models, are compared first in 2D against wind tunnel data to understand their accuracy and limitations. The experimental data for this comparison uses the NACA0015, NACA0030 and NACA4415 airfoils over a variety of cases relevant to wind turbines. Hereafter the models are compared within an aeroelastic code as a part of complete horizontal axis wind turbine simulations for an extreme load case (IEC design load case 1.4), standstill instabilities (IEC design load case 6.2) and the flutter speed to understand the effect of the different dynamic stall models. For the standstill cases the aerodynamic damping provided by each model is compared by reducing the structural damping of the blades to find the point where the blades become unstable. Next the Øye and Risø models are tuned to the wind tunnel data. The results from the 2D comparison are inconclusive with each model showing different strengths and weaknesses. In general the attached flow physics in the Risø and ONERA models improves the fit for the 15% thick airfoils. The ONERA model captures the lift peak the best, although it usually has the largest least squares error to the data due to the drop in lift after this peak being too early and too sharp. Once in the aeroelastic code the ONERA model started to show unphysical behaviour by reducing the deflection in the extreme load case and adding negative damping to the standstill cases. Snel’s model, on the other hand, adds so much damping that the blades remains stable even when the blades have only a small amount of structural damping, which is likely not physical. The Øye and Risø models show similar damping levels and reduce the required structural damping to prevent the standstill instabilities by at least a factor two with respect to no dynamic stall model being used. For the classical flutter analysis the Risø model shows an increase in flutter speed as expected from the implementation of Theodorsen’s theory, while the ONERA model decreases the flutter limit. Overall the Risø model is seen as the best for in an aeroelastic code due to showing better behaviour in the full turbine cases than the ONERA and Snel models. Furthermore, it is superior to the Øye model due to correctly modeling the attached flow physics and improving the drag and moment coefficients.